Nando de Freitas

Profil AI Expert

AI Speciality: 
AI Stochastic
Neural Network
Deep Learning
Machine Learning
Robotic
Current occupation: 
Researcher at Deepind
AI Commitment Rate(%): 
0.00'%'

TwitterID: 
@NandoDF
Tweet Visibility Status: 
Public

Description: 
Nando de Freitas is a researcher in the field of machine learning, and in particular in the subfields of neural networks, Bayesian inference and Bayesian optimization, and deep learning. Now a researcher at Deepmind, he does not hesitate to share the latest advances in the field on social networks. He believes that it is by sharing simulators, code and data that robotics will progress.

Aknowledged by:

Not Available

Les derniers messages de l'Expert:

Tweet list: 

2025-02-08 11:12:17 RT @ego4_d: Ego-Exo4D is the world's largest source of egocentric body + hand pose estimates and eye gaze data exists across the dataset. W…

2025-02-08 11:11:38 RT @svlevine: Scaling laws in deep RL? Turns out that batch size, learning rate, and UTD (update-to-data) for getting the most efficient an…

2025-02-08 11:11:21 RT @docmilanfar: Michael Jordan gave a short, excellent, and provocative talk recently in Paris - here's a few key ideas - It's all just m…

2025-02-08 11:11:10 @JFPuget @docmilanfar Machine learning was downplayed by statistics and even by computer science for a long time but now most of AI is just what we used to call machine learning. Machine learning is no longer in the periphery but at the core.

2025-02-07 11:29:17 I look forward to an AI future where loving, compassionate, kind, empathetic, empowering, cooperative, helpful AIs are realised. AIs that become ubiquitous in our lives and help preserve and expand life, economic wellbeing, intelligence and consciousness — all unique in our known… https://t.co/Nlt4mh6Sgw

2025-02-06 17:28:32 RT @MistralAI: Introducing the all new Le Chat: your ultimate AI sidekick for life and work! Now live on web and mobile! https://t.co/MwRwc…

2025-02-06 17:28:15 RT @rasbt: I just finished writing up my take on reasoning models: https://t.co/QLEd6HOh5l Here, I 1. Discuss the advantages &

2025-02-06 17:22:50 RT @neilzegh: Today we release Hibiki, real-time speech translation that runs on your phone. Adaptive flow without fancy policy, simple tem…

2025-02-04 21:23:58 RT @arankomatsuzaki: Improving Transformer World Models for Data-Efficient RL Presents an approach to model-based RL that achieves a new S…

2025-01-31 07:10:58 RT @DavidDuvenaud: New paper: What happens once AIs make humans obsolete? Even without AIs seeking power, we argue that competitive pressu…

2025-01-31 07:09:02 RT @jaseweston: Introducing EvalPlanner – a method to train a Thinking-LLM-as-a-Judge that learns to generate planning &

2025-01-31 07:07:55 RT @jaseweston: Introducing RIP: Rejecting Instruction Preferences A method to *curate* high quality data, or *create* high quality syn…

2025-01-30 08:46:44 @AndreiDavid @scott_e_reed An these are @scott_e_reed and @ashrewards right now: https://t.co/AjBbZgePP2

2025-01-30 08:40:55 @AndreiDavid @scott_e_reed Let me put it this way. There was a federation (incubator) of small capable teams (startups). This was replaced by the death-star. Deepseek is a rebel startup. This is not a take on the people or companies. It is a take on how R&

2025-01-30 07:34:22 In my experience, every amazing project, alphacode, alphago, alphafold, gato, flamingo, etc, started as a small exploration project. I feel that we acknowledge the people at the end of the project more than the pioneers. In fact, most people probably don’t know who started… https://t.co/eaUzEoRg3h

2025-01-30 07:30:07 This thread by @scott_e_reed, one of the best deep learning researchers in the world, summarises well what many experienced working for industrial AI labs over the last two years: 1. Winner take all politics 2. An erosion of our ability to innovate 3. An erosion of our belief… https://t.co/TzEP2HcuVi https://t.co/iNoVNhHBo5

2025-01-30 07:09:46 @scott_e_reed

2025-01-19 12:09:44 RT @zst96687522: The most insightful paper regarding process reward modeling I've seen recently. [2501.07301] The Lessons of Developing Pr…

2025-01-19 12:06:57 RT @_akhaliq: Kokoro is insane. This AI is a groundbreaking TTS model with just 82M parameters. It outperforms larger models and genera…

2025-01-19 12:05:09 RT @mustafasuleyman: After Ethan's post, I went on a deep dive into this study! I could go on and on about the results but if I had to boil…

2025-01-16 08:23:00 @matthewclifford Andrew Zisserman did the same for Oxford Information Engineering. As a result Oxford was able to attract top faculty and students, many contributing to the UK’s economy. In fact, Phil Blunsom and I are part of an old cohort of international PhD students who were lucky to come… https://t.co/2LJqIfux1k

2025-01-16 07:37:28 @matthewclifford I hope you have a discussion with Demis about removing long noncompetes and garden leaves for AI engineers and scientists, @matthewclifford . Google is master of enforcing these in the UK, and they are terrible for innovation and competition. They put us at a huge disadvantage… https://t.co/wfO13Bd4U0

2025-01-16 07:18:46 RT @li_chengzu: Forget just thinking in words. New Era of Multimodal Reasoning Imagine While Reasoning in Space with MVoT Multimodal…

2025-01-14 00:15:23 RT @omarsar0: Multiagent Finetuning Introduces multiagent finetuning, a novel approach for improving language models through self-improvem…

2025-01-14 00:14:43 RT @lmarena_ai: Exciting news from @CopilotArena! The latest Codestral 25.01 release is now topping the Copilot Arena leaderboard (joint #…

2025-01-14 00:05:56 @pcastr There’s a lot more open code in PyTorch. Community, open culture and fun matter. It’s not just about familiarity. Not to mention many powerful libraries for GPU programming.

2025-01-13 23:39:00 Well done @matthewclifford Please get rid of noncompetes in AI in the UK. California does not allow them and this is why California is more competitive than the UK in AI. Getting rid of noncompetes is the best thing you could do to kickstart the AI industry in the UK, and to… https://t.co/go43Bq8Onc https://t.co/5YLZxEbcHw

2025-01-10 13:39:15 RT @mervenoyann: ByteDance just dropped SA2VA: a new family of vision LMs combining Qwen2VL/InternVL and SAM2 with MIT license The model…

2025-01-10 13:36:58 @neilzegh @IEEEsps Congratulations, well deserved

2025-01-09 22:46:16 RT @arankomatsuzaki: SynthLabs + Stanford presents: Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought…

2025-01-08 21:58:45 RT @omarsar0: Don't do RAG Proposes cache-augmented generation (CAG) to eliminate retrieval latency and minimize retrieval errors. What i…

2025-01-08 21:54:03 RT @FrankRHutter: The data science revolution is getting closer. TabPFN v2 is published in Nature: https://t.co/Ybb15pnZ5P On tabular class…

2025-01-04 15:15:22 Sending all my love to his family and friends.

2025-01-04 15:11:04 Heartbreaking, and kind. Thank you, Felix. Your light will continue to shine. https://t.co/eD2COqkian

2025-01-03 21:35:32 RT @armandjoulin: Today I had the great idea of doing chain of thoughts in the continuous space. I know it's a great idea because @jasewest…

2025-01-02 20:44:11 RT @egrefen: I've had the pleasure and honour of knowing and working with Felix since the first year of his PhD. He was one of our first in…

2024-12-30 12:17:52 @_aidan_clark_ That’s not an opinion, Aidan, but a historical fact. I would not put it as people were scared, but rather that people thought the simpler solution of 1 layer SVMs (or Gaussian process) if regularised well could win. They weren’t stupid, they were aiming for the simplest solution… https://t.co/lPQXJUGvwg

2024-12-29 23:48:01 Nothing like the immensity of nature to give us perspective https://t.co/CvZFp56aW8

2024-12-29 23:39:34 I’m grateful for 2024. I had a brilliant start at @GoogleDeepMind working on Lyria, Veo and Imagen 3. I’m pleased that my teams continued to build on our work and deliver amazing results after I left. By leaving Google, I had the chance to go back to the things I love doing the… https://t.co/awizUUeAAU https://t.co/VEaEBqaRtY

2024-12-29 23:11:06 RT @kchonyc: i like both AI and end-to-end encryption. can i have them all? we examine this question carefully in <

2024-12-27 23:06:40 RT @denny_zhou: Chain of Thought Reasoning without Prompting https://t.co/tUeFzcpn0V (NeurIPS 2024) Chain of thought (CoT) reasoning ≠ C…

2024-12-23 11:58:52 @kchonyc Thanks for the thoughtful analysis and post, @kchonyc — you’ve always been a source of wisdom in our AI community.

2024-12-21 11:05:44 Mental health in AI, is also not only about supporting people in need, but also about changing toxic behaviour. Micro-aggressions, e.g. directly putting people down for no reason, is a commonly displayed behaviour in conferences and certain work environments. It is toxic! Please… https://t.co/hj9qer5omJ https://t.co/kifFmZfw9B

2024-12-21 10:37:49 RT @mmmbchang: AI hit a wall and broke through the wall

2024-12-21 10:33:59 RT @Alibaba_Qwen: Qwen2.5 Technical Report https://t.co/09b9WvA9pY https://t.co/NChJRzsvof

2024-12-19 11:41:23 RT @Dominic1King: We're looking for an outstanding Program Manager to join the new health group at Microsoft AI. This position presents a g…

2024-12-17 15:10:37 @MiaosenWang Congratulations Miaosen! It’s very cool!

2024-12-17 15:10:01 @hyunjik11 Congratulations, Hyunjik!

2024-12-17 15:09:23 @RubenEVillegas Congratulations, Ruben!

2024-12-17 11:56:50 RT @wellingmax: Beautifully said Nando. We should slow down the crazy rat race and enjoy the beautiful work we do without all the unnecessa…

2024-12-17 11:56:14 @balazskegl

2024-12-15 22:12:04 Let us please talk more about mental health in the AI community. I was shocked and reminded of this by the sad and tragic death of this young colleague with so much talent. Many of the people in our community are likely on the spectrum

2024-12-15 21:03:35 RT @BhanuKonepalli: @AravSrinivas https://t.co/1xJHhlSLcc

2024-12-15 20:25:15 RT @ArnaudDoucet1: The slides of my NeurIPS lecture "From Diffusion Models to Schrödinger Bridges - Generative Modeling meets Optimal Trans…

2024-12-15 12:00:29 I agree with Dhruv that we are not running out of data but only human written text. The machines are now generating valuable data, eg all the captions used to train Dalle-3 and Imagen-3. The age of self-training and synthetic data has just begun. Moreover, we are far from… https://t.co/DHmjgKkRZg https://t.co/SJ1iElUmkb

2024-12-13 14:20:44 @blaiseaguera And it also includes the environment in that collective! I absolutely love talking about intelligence with you @blaiseaguera - you’re one of the great thinkers.

2024-12-13 12:53:12 RT @deepseek_ai: DeepSeek-VL2 is here! Our next-gen vision-language model enters the MoE era. DeepSeek-MoE arch + dynamic image tillin…

2024-12-13 01:26:44 RT @ciguleva: Whoever has access to Sora, Runway, Pika, Haiper, Luma, Kling… you name it. Could you please animate this guy? Let's compare…

2024-12-11 15:08:43 @sainingxie @nmboffi @ma_nanye Thank you Likewise.

2024-12-11 14:03:49 @msalbergo @ma_nanye @marikgoldstein @nmboffi @sainingxie Hahaha I hadn’t meant this a a recruitment move but if Nanye and Mark are looking for industry AI jobs I would certainly love to chat with them at MAI. I also highly recommend them to competitors who care about good research in this area, eg @yusufaytar @ArnaudDoucet1 and… https://t.co/YAq9g57hrm

2024-12-11 13:46:11 @nmboffi @ma_nanye Thank you I hadn’t realised that @ma_nanye is and undergrad. Very impressive indeed. Congratulations and thank you again.

2024-12-11 11:30:56 RT @iScienceLuvr: [MASK] is All You Need New paper from CompVis group, introduces a new method called Discrete Interpolants that builds on…

2024-12-11 11:27:40 RT @alan_karthi: Leaps in AI could have transformative benefits for health &

2024-12-07 17:09:18 RT @iScienceLuvr: NVILA: Efficient Frontier Visual Language Models abs: https://t.co/Ggofp491eN NVIDIA introduces NVILA, a family of open…

2024-12-05 21:58:05 RT @sundarpichai: We’re rolling out Veo to Vertex AI in private preview to help businesses generate high-quality video from a text/image pr…

2024-12-05 21:12:58 RT @sedielem: Diffusion models learn useful internal representations of images, but it's somewhat impractical to use them for feature extra…

2024-12-05 21:10:17 It’s remarkable how many videos have already been generated by a single small startup. Very soon the number of generated videos will surpass the number of videos created by the human race. It remains for AI to make a masterpiece though. This is true in music too. https://t.co/eplQxnhQh7

2024-12-05 17:01:53 RT @jparkerholder: Introducing Genie 2 - our most capable large-scale foundation world model, which can generate a diverse array of cons…

2024-12-05 17:01:45 @jparkerholder Congratulations @jparkerholder and team. Great progress!

2024-12-03 23:12:19 RT @MitoPsychoBio: Superb video of the mitochondrial community in a single cell—hundreds of squiggly microorganisms sensing, processing, an…

2024-12-03 22:31:52 RT @mtschannen: Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained v…

2024-12-03 22:30:05 RT @Dominic1King: Very excited to be joining Microsoft AI. Health is a critical use case for AI. Patients and the public continue to face…

2024-12-03 22:30:01 @Dominic1King Welcome to Microsoft AI, Dom. Really excited to have brilliant people like you harnessing AI to improve healthcare and people’s lives.

2024-12-03 22:20:23 RT @RuiqiGao: A common question nowadays: Which is better, diffusion or flow matching? Our answer: They’re two sides of the same coin. W…

2024-12-01 11:45:34 @boazbaraktcs Precisely, Boaz. The implications of these policies, which employees don’t challenge for fear of retaliation, are profound. We must question how AI is being developed, and not forget that corporations are just utilitarian entities.

2024-12-01 11:40:29 @wightmanr True north I’m proud of Canada for their stance on this.

2024-12-01 11:38:31 @LucianaBenotti Hola Luciana, Si también. Pero al principio uno por lo menos tiene otras opciones usualmente. Es más difícil cuando te cambian el contrato y no tienes la opción de otros trabajos.

2024-11-30 22:48:58 @eisokant @poolsideai Thank you. This is the edited post. https://t.co/Tb3bL8KvuJ

2024-11-30 22:48:18 @tunguz Edited version. Thanks. They also reach out to me, many tearful. Yet people are terrified to speak up. https://t.co/Tb3bL8KvuJ

2024-11-29 20:57:24 RT @_akhaliq: Qwen-Agent Qwen-Agent is a framework for developing LLM applications based on the instruction following, tool usage, plannin…

2024-11-27 16:35:59 Correction, LLM not VLM data yet. @allen_ai when should we expect the image data of MOLMo? Loved the way you collected it with audio

2024-11-27 09:09:04 The OLMo 2 VLM effort is a great example of the value of good data acquisition and curation — Data is the most precious commodity in AI, other than people and compute. While most of open source focuses on models, most of the big recent innovations by industrial labs are data… https://t.co/r8Ji0TQsf4 https://t.co/h2dqVWHCGk

2024-11-26 08:37:46 RT @prmshra: A friend asked me to explain DNA, RNA, and epigenetics. he said that others had tried before, but it didn’t click for him. I…

2024-11-26 07:53:54 RT @_akhaliq: O1 Replication Journey -- Part 2: Surpassing O1-preview through Simple Distillation, Big Progress or Bitter Lesson? https://t…

2024-11-23 20:41:09 @AravSrinivas @perplexity_ai @denisyarats It really is very good. It is providing me with a lot of inspiration, and sorry I’m in one of the competing teams I was also inspired by your interview with Lex Fridman. It has been a pleasure to watch you growing over the last 10 years and see you become one of the greatest… https://t.co/tt9ZGkUgYn

2024-11-23 20:23:54 @pracwis @sebkrier Yes, the public and their many elected governing bodies have done a good job in managing the largest arsenals of nuclear weapons in the world. It is legislation and democracy at play. Even authoritarian states have to comply with legislation and institutions. It’s not perfect,… https://t.co/04Ew0cL9h3

2024-11-23 11:32:48 @sebkrier Hi Seb, since you work in policy dev and strategy at GDM, maybe you could explore and analyse the following examples: 1. AI for surveillance: Let the people in each democratic country vote on laws about this. Encourage transparency. After all, “Democracy is the worst form of… https://t.co/YBD59CzW4s

2024-11-23 11:14:17 RT @omarsar0: Nice paper from Alibaba on building open reasoning models. They propose Marco-o1 which is a reasoning model built for open-…

2024-11-22 12:59:15 The OpenAI letters: https://t.co/aOK7hE6N11 Some of what is said here is absolutely shocking. The politics, hysteria, incompetence, power hunger, gaslighting, etc are beyond any HBO show. I was a leading researcher at DeepMind at the time reporting to Demis. Most of what is… https://t.co/6Lc9hmN3kl

2024-11-21 21:49:15 RT @bfl_ml: Today, we are excited to release FLUX.1 Tools, a suite of models designed to add control and steerability to our base text-to-i…

2024-11-21 21:47:40 RT @deepseek_ai: Introducint JanusFlow: harmonizing autoregressive LLMs with rectified flow! By adopting the best practices in both fie…

2024-11-21 21:47:05 RT @deepseek_ai: DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power! o1-preview-level performance on AIME &

2024-11-21 21:31:16 RT @PaglieriDavide: Tired of saturated benchmarks? Want scope for a significant leap in capabilities? Introducing BALROG: a Benchmark fo…

2024-11-20 23:24:23 I’m loving @perplexity_ai. It is likely the best search experience in the world. Well done @AravSrinivas @denisyarats and team. Amazing UI. https://t.co/ybqD1WFR7V

2024-11-20 00:11:47 RT @MistralAI: We also released Pixtral Large, a new SOTA vision model. https://t.co/cE7Drzasvv

2024-11-20 00:05:53 RT @Alibaba_Qwen: After the release of Qwen2.5, we heard the community’s demand for processing longer contexts. https://t.co/zu2mOB9CQm…

2024-11-19 08:45:20 RT @WorldEverett: Midjourney + Runway + Hailuo AI + Udio: "Childhood Dream" (AI music video) My favorite toy as a kid was LEGO. The idea…

2024-11-19 08:11:33 RT @omarsar0: The Dawn of GUI Agent Explores Claude 3.5 computer use capabilities across different domains and software. They also provi…

2024-11-17 22:48:46 And this is what intelligence looks like: odour perception, prediction aka dream generation (maybe visual, definitely motor), complex action. It’s a never ending process of association. Intelligence is as much about the cat as it is about what is going on in its environment: The… https://t.co/veSz8m4nSD https://t.co/4EHpu5wIzk

2024-11-17 21:42:13 @docmilanfar Haha here is the rest. Not as good. Sometimes it’s very good, to the point I feel uncomfortable about posting it and other times a lot less funny. Still remarkable that it can work out of the box. And of course only fair if others decide to roast me https://t.co/fvOtoe9MPd

2024-11-16 20:39:15 @docmilanfar https://t.co/w0J9DwRAmT

2024-11-16 20:28:49 @sirbayes Search for my friend Kevin P Murphy and write a roast about him https://t.co/BDsYUt1DlN

2024-11-16 12:01:10 It is truly remarkable that we now have machines capable of mapping my speech to a text query “Search for my friend Scott Reed, who worked with me at DeepMind, and write a short roast about him” and are able to understand that I’m asking it to search the web, searches the web,… https://t.co/NI3jS4sWmQ https://t.co/7gTs7dmOqc

2024-11-16 11:50:38 @danyallah_ Related, though I haven’t researched this for a while, https://t.co/XHQgeOttSy also https://t.co/bB4z9YrKs9 and https://t.co/DS3rUrFLEs

2024-11-16 00:09:03 RT @sedielem: Big fan of straightforward ideas that help to free us from the tyranny of the grid! Off-the-grid ideas often end up being too…

2024-11-14 21:24:10 RT @keenanisalive: We often think of an "equilibrium" as something standing still, like a scale in perfect balance. But many equilibria ar…

2024-11-14 21:23:21 RT @rsalakhu: My talk on AI Agents is online: https://t.co/Kk7BS40rZl

2024-11-14 21:17:53 @doomie Priceless

2024-11-13 22:44:01 RT @omarsar0: Lots of great tips and insights for devs building RAG systems. "Our experiments suggest that models that can retrieve a high…

2024-11-13 22:43:07 RT @_akhaliq: Nvidia presents Add-it Training-Free Object Insertion in Images With Pretrained Diffusion Models https://t.co/IGYItL9E8D

2024-11-12 23:49:32 RT @brandondamos: Flows and transport methods are widely used to connect one distribution to another. What about going up one level to tran…

2024-11-12 23:48:41 RT @scott_e_reed: Congrats! Cool to see that latent actions are not only useful for interactive world models (as in genie) but also as targ…

2024-11-12 22:33:01 RT @scott_e_reed: Very cool idea: make the diffusion policy denoising process part of the MDP and train the whole thing with PPO.

2024-11-11 22:57:25 and this is how ChatGPT sees me, “Based on what you know about me, draw a picture of what you think my life looks like” @OpenAIDevs I kind of like it https://t.co/XtEaIhWzoa

2024-11-11 22:49:09 RT @mustafasuleyman: Fascinating and important work: AI2BMD holds great potential for understanding the mystery of biological systems and d…

2024-11-10 20:45:26 @Noahpinion @polynoamial A common practice in industry is to train very large models and then distill them to smaller ones for deployment. So scaling is a necessary optimisation step. Another common practice to increase capacity is MoEs. So scaling is happening and models are getting better over time.… https://t.co/MpeywTTh3n

2024-11-10 19:59:04 RT @akyurekekin: Why do we treat train and test times so differently? Why is one “training” and the other “in-context learning”? Just tak…

2024-11-10 12:17:08 What is being scaled? is a brilliant question, @tdietterich I once posted: “It’s all about scale now! The Game is Over! It’s about making these models bigger, safer, compute efficient, faster at sampling, smarter memory, more modalities, INNOVATIVE DATA, on/offline” DATA was… https://t.co/DRl4bfgghm https://t.co/4jFR9ytJcj

2024-11-09 00:29:00 RT @scott_e_reed: https://t.co/1b4ArDfsPg

2024-11-09 00:28:56 @scott_e_reed Thanks for the great insights into this paper, @scott_e_reed

2024-11-09 00:17:15 RT @jaseweston: Self-Consistency Preference Optimization (ScPO) - New self-training method without human labels - learn to make the mode…

2024-11-07 09:23:13 @SuryaGanguli Hi Surya, I still feel +. The dream lives on. One day we’ll be 1 world prosperous, caring, free, living in peace. We just need to keep working towards it. I hope you’re doing well, it’s been a while. Warm wishes

2024-11-06 23:50:01 @SuryaGanguli There’s plenty of innovation here, but you’ll certainly pay lower taxes there. No place is perfect, but we’re happy here.

2024-11-06 20:36:00 RT @satyanadella: Congratulations President Trump, we’re looking forward to engaging with you and your administration to drive innovation f…

2024-11-06 07:30:27 RT @bingyikang: Curious whether video generation models (like #SORA) qualify as world models? We conduct a systematic study to answer this…

2024-11-06 00:47:07 @roydanroy

2024-11-06 00:45:40 RT @roydanroy: Deep learning.

2024-11-06 00:23:06 RT @j_foerst: Currently Deep RL is going through an imagenet moment and very few people are aware. This has major implications for RL appli…

2024-11-04 23:44:01 RT @Rainmaker1973: This is from The Tonight Show with Johnny Carson aired on May 20th, 1977. Carl Sagan says something very important,…

2024-11-02 08:18:32 RT @minchoi: This is wild. Runway just dropped Advanced Camera Control for Gen-3 Alpha Turbo. Now you can choose the direction and intens…

2024-11-01 19:24:55 This is finally beginning to demonstrate the power of LLMs to organize the world's information and make it universally accessible and useful congratulations @sama and @OpenAI team! https://t.co/l9VzZCgkfm

2024-11-01 19:20:14 @danfei_xu Great effort @danfei_xu Very cool

2024-11-01 07:06:18 RT @historyinmemes: Steve Wozniak's Apple I (1976). https://t.co/146OHAoFKJ

2024-11-01 07:01:25 RT @sama: searching for a chrome extension is not easy, so here is the link: https://t.co/7tYkmMgfjr

2024-10-31 21:53:21 RT @OpenAI: Introducing ChatGPT search ChatGPT can now search the web in a much better way than before so you get fast, timely answers…

2024-10-31 00:14:35 RT @AdiSimhi: LLMs often "hallucinate". But not all hallucinations are the same! This paper reveals two distinct types: (1) due to lack of…

2024-10-31 00:13:27 @zalanborsos Amazing to see this! Congratulations, Zalan &

2024-10-31 00:12:19 RT @zalanborsos: Our brief overview of the audio generation model behind NotebookLM Audio Overviews and Illuminate.

2024-10-29 23:30:55 @emiel_hoogeboom @RuiqiGao @dpkingma Thanks!

2024-10-29 23:26:53 RT @HaiperGenAI: Introducing Haiper 2.0: Text-to-Image Like Never Before! Unleash your creativity with sharper, more realistic visuals…

2024-10-28 09:00:15 @yudapearl GPT4o (not the best, which is o1) already delivers some of the arguments about your statement for us to examine. Not that different from Twitter! GPT-4o: The statement raises a thought-provoking point about how large language models (LLMs) like ChatGPT-4 interact with causal… https://t.co/LPwjjHNMqV

2024-10-27 22:27:35 RT @reach_vb: Wow! Meta dropped an open NotebookLM recipe: NotebookLlama It uses L3.2 1B/ 3B for pre-processing the PDF, L3.1 70B for Tr…

2024-10-27 21:52:53 RT @yukez: Another key result from my lab in leveraging human-centered data sources for humanoid robots — this time, human motion captures.…

2024-10-26 20:34:25 @jeffclune I think we’re more in a situation where multimodal LLMs are useful tools to create diagrams, improve the write up, etc. That is, there’s potential for the quality of papers to improve thanks to these “assistants”. It also makes academia more accessible to people whose first… https://t.co/6RLrPElnQq

2024-10-26 20:25:02 @sirbayes I wonder how many people already do it without admitting it One thing that makes this technology interesting is the unreported uses.

2024-10-26 19:03:15 @yudapearl I took the liberty of asking your question. The answer appears below. I agree with all your comments regarding this unusual “student”: too easy to nudge, doesn’t get to the heart of things, and definitely is incapable of creating and advancing knowledge at large. It has many… https://t.co/TjuYw08GQW

2024-10-26 18:36:05 @jchencxh I didn’t think either until my little daughter found one. I’ve been searching since then. She lost hers, so I gave one of mine to her

2024-10-26 18:32:34 @eliasbareinboim I think one day anyone could go to something like ChatGPT to learn about new topics. It’s already starting to happen, and I’m excited about it, especially because it democratises knowledge and education, in a world where most still lack proper access to it. My tweets are about… https://t.co/AhMHAIz0PU

2024-10-24 23:46:31 @MelMitchell1 Hahaha, I was hoping it would build on https://t.co/xbw9D14Dkb and derive new insights, but not quite …. It’ll take a lot more work but I did find its summaries to be super impressive

2024-10-24 23:41:59 @_lukaemon Cool. Thanks so much for sharing

2024-10-24 23:40:18 @MelMitchell1 How close?

2024-10-24 23:39:09 @tdietterich You could argue that the organism seeks order in an environment of growing entropy. So exploration is forced upon the agent by the environment. I feel Intelligence is meaningless without the existence of an environment. Keeping old genes could be about survival in very long… https://t.co/9hekdD996v

2024-10-24 23:21:39 How do we measure order? Is Entropy enough or do we need measures of complexity? GPT-4o: … ### 2. **Complexity as a Measure of Order: Complexity captures not just the degree of randomness or order but also the level of **organization, structure, and function** within a system.… https://t.co/Q2aIwSiC0K

2024-10-22 22:32:58 RT @AnthropicAI: Introducing an upgraded Claude 3.5 Sonnet, and a new model, Claude 3.5 Haiku. We’re also introducing a new capability in b…

2024-10-22 22:20:28 RT @AnthropicAI: We've built an API that allows Claude to perceive and interact with computer interfaces. This API enables Claude to trans…

2024-10-22 22:19:12 RT @perplexity_ai: Pro Search is now more powerful. Introducing Reasoning Mode! Challenge your own curiosity. Ask multi-layered questions.…

2024-10-22 22:18:40 RT @ideogram_ai: Today, we’re introducing Ideogram Canvas, an infinite creative board for organizing, generating, editing, and combining im…

2024-10-22 22:18:05 RT @runwayml: Introducing, Act-One. A new way to generate expressive character performances inside Gen-3 Alpha using a single driving video…

2024-10-20 15:12:20 RT @hardmaru: NVIDIA’s Llama-3.1-Nemotron-70B-Instruct-HF can be used on HuggingChat https://t.co/lMcXKpXRmU https://t.co/07ntkQyGLw

2024-10-20 15:11:52 RT @NVIDIAAIDev: Our Llama-3.1-Nemotron-70B-Instruct model is a leading model on the Arena Hard benchmark (85) from @lmarena_ai. Arena…

2024-10-20 13:05:55 RT @GabiImmelman: @NandoDF I’m a South African and there’s something more disturbing about the poverty in San Francisco and it’s somewaht m…

2024-10-20 13:04:21 @GabiImmelman That’s an excellent point, Gabi. So true. As a South African, I felt the same.

2024-10-20 12:59:44 RT @PetarV_93: Round and Round we Go! Rotary Positional Encodings (RoPE) are a common staple of frontier LLMs. _Why_ do they work so we…

2024-10-18 16:46:27 RT @omarsar0: LLMs Can Learn About Themselves by Introspection This paper reports that LLMs can acquire knowledge through introspection th…

2024-10-18 16:41:40 RT @AIatMeta: As detailed in the Meta Movie Gen technical report, today we’re open sourcing Movie Gen Bench: two new media generation bench…

2024-10-18 16:41:01 RT @SamuelMLSmith: I ambushed a theory workshop with a tutorial on scaling LLMs: https://t.co/HJtAiZUHFm Covers transformers, a simple mod…

2024-10-18 14:56:20 RT @lmarena_ai: Introducing Copilot Arena - Interactive coding evaluation in the wild. Our extension lets you test top models for free, ri…

2024-10-18 06:35:22 RT @robot_trainer: Cool. They're getting the job done. Combination of classical decomposition (eg mapping + navigation) and e2e nets (perfo…

2024-10-13 15:55:47 @anilkseth @DarioAmodei I agree with your point about self-awareness and awareness. I think we can be aware of many things: our actions or interventions leading to causal statements P(.|do(a)), our bodies either consciously or unconsciously, our attention as Michael Graziano argues. We can train models… https://t.co/ZKUlT0xFZV

2024-10-13 11:54:29 @ilyasut you’re a very clear thinker who I believe thinks about this a lot too. What are your thoughts on this question of awareness and safety.

2024-10-13 11:40:30 @boazbaraktcs Loved this

2024-10-13 11:40:02 RT @hardmaru: Diffusion models seem to be particularly well suited for game engine world models. In the future, a neural simulator agent ca…

2024-10-13 11:38:08 @JeremyDanielFox I agree wholeheartedly.

2024-10-12 22:17:18 RT @slow_developer: Ilya Sutskever had a conversation last year with Jensen Huang but still, that made me rethink the idea of 'Predicting t…

2024-10-12 22:08:48 RT @JeffDean: My @Google colleague and longtime @UCBerkeley faculty member David Patterson has a great essay out in this month's Communica…

2024-10-11 06:04:04 RT @CollinRugg: NEW: Elon Musk introduces an army of Optimus robots, says people will be able to buy them to complete tasks. Epic. Musk…

2024-10-11 06:01:49 RT @arankomatsuzaki: Scaling Laws For Diffusion Transformers https://t.co/Uw3E7wzN9q https://t.co/voumdosUSO

2024-10-10 20:26:08 @kellerjordan0 @bozavlado I’m enjoying watching this collaboration

2024-10-10 20:23:07 RT @_akhaliq: New Kling, Runway, Luma competitor? Open text and image video generation model Pyramidal Flow Matching for Efficient Video…

2024-10-10 18:11:06 RT @OpenAI: We’re releasing a new benchmark, MLE-bench, to measure how well AI agents perform at machine learning engineering. The benchmar…

2024-10-10 18:10:31 RT @ArnaudDoucet1: Yet another super nice paper by Hyvarinen, how to implement Langevin when only having access to the score of a noised ve…

2024-10-10 06:10:37 @clemenslm Congratulations, Clemens!

2024-10-08 22:13:25 @kchonyc How did you manage to pass?!

2024-10-08 22:12:31 RT @tsarnick: Geoffrey Hinton says he is "flabbergasted" about being awarded the Nobel Prize in Physics and he believes AI will exceed peop…

2024-10-08 22:03:53 RT @satyanadella: Our long-standing partnership with NVIDIA and deep innovation continues to lead the industry, powering the most sophistic…

2024-10-08 21:15:41 Highly deserved recognition. John Hopfield and @geoffreyhinton inspired an entire generation of researchers in the pursuit of understanding how brains work, intelligence and thinking. They provided unparalleled and extraordinary thought leadership, and made us dream that it was… https://t.co/xSF3VcIQ1W https://t.co/Ey4wkZ3GXE

2024-10-07 23:15:33 RT @j_foerst: Meet an incredibly simple and effective method for compute-only self-improvement: simply add a checklist to evaluate answer q…

2024-10-01 21:50:07 RT @mustafasuleyman: Today we’re launching our new Copilot experience. I truly believe we can deliver a calmer, more helpful and supportive…

2024-10-01 21:49:03 RT @maheeldabarera: @NandoDF lecture on #GaussianProcesses is the best lecture I have heard on the topic. Took me 6 years to understand it…

2024-09-30 22:35:58 RT @skirano: o1-engineer is here! A coding assistant built from the ground up to leverage o1 reasoning capabilities. It can create and…

2024-09-30 22:16:44 RT @rasbt: This paper is actually a nice find! "Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models" (http…

2024-09-30 07:19:01 RT @polynoamial: @OpenAI o1 is trained with RL to “think” before responding via a private chain of thought. The longer it thinks, the bette…

2024-09-28 22:33:58 RT @lmsysorg: Exciting update from Vision Chatbot Arena! We’ve gathered over 6K new votes for the latest open vision models (Qwen2, Llama…

2024-09-28 22:31:33 RT @AlphaSignalAI: Anthropic just reduced the error rate of RAGs by 67% using a ridiculously simple method. They add important context to…

2024-09-28 12:10:53 RT @dawnsongtweets: Join us for 4th lecture on Enterprise trends and use cases in LLM Agents in our MOOC, Burak Gokturk @Google, 3:10pm Sep…

2024-09-28 12:08:49 Last week we saw many releases of powerful multimodal understanding AI models, including Llama 3.2, NVidia’s NVLM, Molmo and many more. All great works. But here are my reflections on what Open-Source needs to do to advance AI: 1. There exist many public datasets to pretrain and… https://t.co/rQws2lKI3l

2024-09-28 11:51:08 I guess this is the equivalent of slicing a data-centre full of GPU racks! It’s interesting how we now have millions of “artificial brains” expressed in software in virtual machines, or docker containers. The hardware — phones, laptops, local clusters, massive data-centres all… https://t.co/fHdZSoKwIc https://t.co/h2JChZvmU7

2024-09-27 06:00:49 RT @udiomusic: Introducing the Udio Lyric Editor, now available for all users. Generate random or prompt-based lyrics Weave in your…

2024-09-26 22:31:59 @OxxoTweets @anikembhavi @DrJimFan We owe you. Thank you

2024-09-26 22:31:22 @anikembhavi @DrJimFan Thanks for suggesting. Molmo 72B is a pretty good model.

2024-09-26 22:28:42 @mattdeitke Congratulations. The results are brilliant

2024-09-26 22:24:44 @Muennighoff Nice work.

2024-09-24 06:39:29 RT @rasbt: Wondering how GPT and Llama compare under the hood? I built a step-by-step code notebook to break down the key differences: http…

2024-09-24 06:37:10 RT @jaseweston: New paper! Backtracking Improves Generation Safety - We train LLMs with DPO to output a RESET token mid-generation if it…

2024-09-23 22:24:35 RT @polynoamial: o1-preview is pretty good at planning https://t.co/PPW35lQ2vD

2024-09-23 22:23:36 RT @minchoi: OpenAI o1 is wild. It's only in "Preview" without vision and people are already doing incredible things and enhancing their w…

2024-09-23 22:21:35 @sama Well said.

2024-09-21 12:56:31 RT @polynoamial: .@OpenAI is hiring ML engineers for a new multi-agent research team! We view multi-agent as a path to even better AI reaso…

2024-09-20 23:10:53 RT @AdeenaY8: InfiMM-WebMath-40B an open multimodal dataset designed for complex mathematical reasoning, released by @ByteDanceOSS Datas…

2024-09-20 23:09:15 RT @ProfBuehlerMIT: Introducing LifeGPT, showing that LLMs can simulate complex, Turing-complete systems like Conway's Game of Life with ne…

2024-09-20 23:04:18 RT @OpenAI: Some of our researchers behind OpenAI o1 https://t.co/XnMx9vY2J2

2024-09-20 23:03:11 @UBC_CS @UBC_NLP @CAIDA_UBC @careninigiusepp @ubcscience Bravo

2024-09-20 23:01:44 The @OpenAI o1 models represent one of the smartest advances in AI in a long time. Having just joined @Microsoft AI, one of the things I really look forward to is being able to contribute to some of these fruitful ideas to advance OpenAI’s mission. The opportunity to work… https://t.co/PDoqwU5AD8 https://t.co/2dRoNXDl6K

2024-09-19 06:09:59 RT @iScienceLuvr: Playground v3: Improving Text-to-Image Alignment with Deep-Fusion Large Language Models abs: https://t.co/3EZbWVWOfu Ne…

2024-09-18 20:46:35 RT @_weiping: Introducing NVLM 1.0, a family of frontier-class multimodal LLMs that achieve state-of-the-art results on vision-language tas…

2024-09-18 20:45:56 RT @_akhaliq: Fine-Tuning Image-Conditional Diffusion Models is Easier than You Think discuss: https://t.co/s5QqGZZFAa Recent work showed…

2024-09-18 20:39:41 RT @LiuXub: I'm excited to introduce the Source-Disentangled Neural Audio Codec (SD-Codec), a new codec model that can disentangle arbitrar…

2024-09-18 20:39:11 RT @honualx: Meet Moshiko and Moshika, the open source Moshi models . Moshi is a 7B text-audio model, capable of doing full-duplex conver…

2024-09-16 20:53:07 @mustafasuleyman Thanks @mustafasuleyman. I’m looking forward to working with you again and getting to know the team better. First day at work was Amazingly cool new ideas! Loved it.

2024-09-16 20:43:19 RT @iScienceLuvr: GitHub has incorporated gpt-o1-preview in Copilot https://t.co/5sYQvs54Ib "The results highlight how o1-preview’s reaso…

2024-09-16 17:11:22 @MakkarNik @Microsoft Thanks

2024-09-16 17:10:45 @gowthami_s @Microsoft Hi Gowthami, please check your inbox.

2024-09-16 16:09:01 I’ve joined @Microsoft AI to advance the frontier of large scale multimodal AI research and to build products for people to achieve meaningful goals and dreams. The MAI team is small, but well resourced and ambitious. We are now looking for exceptional ICs, who like to ship. If… https://t.co/y1sAjgvWht https://t.co/oofQSePycj

2024-09-14 11:41:11 @avishkar58 @GoogleDeepMind I very much look forward to continuing to participate in the @DeepIndaba. Helping you, Shakir and Ulrich start it was a very enriching experience. It gave me so much, and it keeps giving. The deep learning Indaba became a model for @Khipu_AI and many other amazing education… https://t.co/UivZEu9ruN

2024-09-14 11:34:02 RT @avishkar58: @NandoDF @GoogleDeepMind We will miss you at DeepMind, Nando! Thank you for always setting such a great example for the res…

2024-09-14 11:29:44 @wellingmax @AxcanNathan It depends on the startup

2024-09-14 11:28:16 @wellingmax I would love to hear the opinion of game theorists and geopolitical experts on this. I worry that this California choice will impact all of us, and that California, although amazing, is not always the best at solving problems (eg dire state of homelessness in SF). I find it hard… https://t.co/iTXMaRrf8W

2024-09-14 09:45:54 @drfeifei @jcjohnss Congrats Fei-Fei, it is a great mission.

2024-09-10 22:14:03 RT @blaiseaguera: Thank you @alokjha and @TheEconomist for having me on Babbage!

2024-09-10 16:05:48 RT @yoavgo: is instructing an LLM to "not hallucinate" absurd? i used to think it is, but upon some reflection, i think it really isn't, an…

2024-09-10 16:04:09 RT @joshim5: We’re excited to introduce @ChaiDiscovery and release Chai-1, a foundation model for molecular structure prediction that perfo…

2024-09-10 16:02:53 RT @DrJimFan: It is *incredibly* easy to game the LLM benchmarks. Training on test set is for the rookies. Here're some tricks to practice…

2024-09-10 10:51:23 RT @stanley_h_chan: Tutorial on Diffusion Models for Imaging and Vision 2nd edition is up on arXiv https://t.co/fWv111nG5M (51 pages -->

2024-09-07 16:00:18 RT @osanseviero: DeepSeek 2.5 is out! A powerful MOE with 238B params with 160 experts and 16B active params Chat and code capabilities…

2024-09-07 15:59:27 RT @WiMLworkshop: Ready to boost your ML/AI career or share your expertise? Join our Women in Machine Learning Mentoring Program! Sign u…

2024-09-07 15:55:12 RT @rasbt: Just added a multi-head attention implementation for Einstein summation enthusiasts to my collection: https://t.co/eFXjxJ5Sev Wh…

2024-09-07 11:31:52 RT @ericjang11: Really enjoyed this @karpathy interview. I think people disagree with his takes because he's slightly ahead of the curve an…

2024-09-07 09:53:50 RT @dawnsongtweets: Large Language Model Agents is the next frontier. Really excited to announce our Berkeley course on LLM Agents, also av…

2024-08-23 09:33:44 @sedielem Waiting eagerly! Thanks

2024-08-23 09:18:27 RT @sedielem: It's so much easier to tweet low-effort memes which assert that diffusion is just autoregression in frequency space, than it…

2024-08-23 09:18:25 @sedielem The blogpost: https://t.co/hMfSXpLdlT.

2024-08-23 09:13:12 RT @JeffDean: Vizier is a tool used in tens of thousands of places at Google for a huge variety of different black box optimization tasks,…

2024-08-23 09:10:17 RT @sedielem: Think you understand classifier-free diffusion guidance? Think again! These two papers beg to differ https://t.co/ll9dph8Th…

2024-08-21 16:28:02 RT @rohanpaul_ai: Wild idea in this paper Current LLMs are stateless between tokens which lead to many many problems requiring reasoning…

2024-08-21 16:27:12 RT @minchoi: Where is Sora? Creatives have been working with OpenAI Sora and dropping these banger AI videos, rather quietly. 11 wild one…

2024-08-21 16:25:26 RT @osanseviero: Microsoft just 3 new models - Phi 3.5 mini instruct (3.8B, 128k context length) - Phi 3.5 MoE (42B-A6.6B, 128k context…

2024-08-21 16:23:37 RT @arankomatsuzaki: Meta presents Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model - Can generate images…

2024-08-20 20:14:49 RT @iScienceLuvr: LongVILA: Scaling Long-Context Visual Language Models for Long Videos abs: https://t.co/zrbENzeBed code: https://t.co/G8…

2024-08-19 08:06:46 RT @_akhaliq: JPEG-LM LLMs as Image Generators with Canonical Codec Representations discuss: https://t.co/GV0AL9ajLp Recent work in imag…

2024-08-19 08:06:12 RT @iScienceLuvr: xGen-MM (BLIP-3): A Family of Open Large Multimodal Models abs: https://t.co/b0lt7bfm8f model: https://t.co/ksZSwyiNGx…

2024-08-19 08:04:17 RT @arankomatsuzaki: Automated Design of Agentic Systems Presents Meta Agent Search to demonstrate that we can use agents to invent novel…

2024-08-18 22:37:00 RT @hyhieu226: New tutorial on WGMMA (WarpGroup Matrix Multiplication and Accumulation) https://t.co/mmwphf9Zxb If you have run PyTorc…

2024-08-18 19:01:19 RT @cjmaddison: @pfau https://t.co/C8WuQmvcUL https://t.co/VMqcc7T4IY

2024-08-16 15:21:23 RT @_akhaliq: Generative Photomontage discussion: https://t.co/8jlXRjYSGm Text-to-image models are powerful tools for image creation. How…

2024-08-16 15:19:09 RT @arankomatsuzaki: DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search Sign…

2024-08-16 14:52:55 https://t.co/xdOtgBLlNK (a ref for history inclined folks). See also How Disney Built America https://t.co/J2DtkOqXjs

2024-08-16 14:28:33 RT @BenjaminDEKR: Imagen 3 ->

2024-08-16 14:13:26 @avdnoord Great work, Aaron. Congratulations!

2024-08-14 12:56:25 RT @lmsysorg: Exciting Update from Chatbot Arena! The latest @OpenAI ChatGPT-4o (20240808) API has been tested under "anonymous-chatbot" f…

2024-08-14 08:26:26 RT @_akhaliq: ControlNeXt authors are active in the discussion section on Hugging Face reach out to them here: https://t.co/ZCCFddBpWL

2024-08-14 08:23:00 RT @CerebrasSystems: Cerebras Co-Founder Deconstructs NVIDIA Blackwell Delays From intricate interposer designs to alignment issues and th…

2024-08-14 08:21:36 RT @lmsysorg: Woah, another exciting update from Chatbot Arena The results for @xAI’s sus-column-r (Grok 2 early version) are now publ…

2024-08-14 08:20:13 RT @iamtrask: (obviously) transformers are the most in-demand AI technique this *from scratch* transformers tutorial is so complete and #…

2024-08-10 02:14:18 Andrew Gelman, one of the world’s most renowned statisticians, on the attempts of dictator Maduro to remain in power in Venezuela. Maduro continues to cause needless suffering to millions of people for his own selfish, criminal, financial gain. @MariaCorinaYA @elonmusk @JMilei https://t.co/AWtUZGe9WT

2024-08-04 22:37:52 RT @JeffDean: An overview of @GoogleAI work showing one way that AI is helping reduce worldwide emissions, by better timing stop lights to…

2024-08-04 01:16:42 RT @ZelenskyyUa: Worrying reports of Russian Wagner mercenaries being spotted in Venezuela alongside government forces. Wherever these thug…

2024-08-03 02:01:54 RT @MariaCorinaYA: Mis queridos y valientes venezolanos, como cada vez que nos hemos levantado, mañana sábado nos crecemos y en familia, en…

2024-08-03 02:01:25 RT @sedielem: I gave a 1-hour talk about generative modelling at the EEML 2024 summer school last month. It's mostly an intuitive look at…

2024-08-03 02:00:43 RT @LightningAI: The creators of Stable Diffusion just released FLUX.1, a powerful new open source image generation model Try it out wit…

2024-08-03 01:59:10 RT @MariaCorinaYA: IMPORTANTE‼ Ofrecemos al mundo la verdad: los resultados detallados de nuestra victoria, que el CNE no presentó en el…

2024-08-02 06:08:15 RT @jiayq: People often ask why prices like $2.8/m token for Llama 405B, while being super fast, are still profitable at @LeptonAI. We've e…

2024-08-02 05:52:52 RT @character_ai: Thrilled to share that we're open sourcing our innovative approach to prompt design! Discover how Prompt Poet is revoluti…

2024-08-02 03:35:54 RT @gil2rok: @LightningAI @ludiXIVwinkler @phillip_lippe Main benefits I enjoy from PyTorch lightning are (1) helps you structure your DL c…

2024-08-02 03:26:18 RT @MLCommons: @MLCommons #AlgoPerf results are in! $50K prize competition yielded 28% faster neural net training with non-diagonal preco…

2024-08-02 03:24:15 @JeffDean @lmsysorg Congratulations Jeff &

2024-07-24 08:33:32 RT @ylecun: BOOM Llama 3.1 is out 405B, 70B, 8B versions. Main takeaways: 1. 405B performance is on par with the best closed models. 2…

2024-07-23 22:04:02 What a stimulating day at #ICML2024! No better way to end it than by having dinner with a dear old friend @j_foerst, enjoying local food, and pondering on: “The transition from non-life to life has never been observed experimentally”. That’s why AI/ML conferences are so amazing… https://t.co/EgwqC9OVWL https://t.co/irABWvbY3L

2024-07-23 21:51:43 RT @chinasza: One of my current research projects is focused on quantifying AI progress in Africa and it’s sad that all prominent AI/ML con…

2024-07-23 21:46:13 @kofiemeritus @icmlconf I think it is more nuanced. It is true that visas exclude people, and sadly even among African countries visas are an issue. There is also bias in the sense that these meetings mostly occur in Europe and north America. It is costly and demanding to travel to these conferences… https://t.co/i4VY0MvMsc

2024-07-23 21:28:55 @4bsolu7 Vienna

2024-07-23 10:15:13 Trading and investment conference, or AI conference? Have a guess https://t.co/2O3zJ0rOQz

2024-07-23 08:39:09 Generating interactive environments https://t.co/ir4i7ijqGk

2024-07-23 07:39:33 https://t.co/lgcxoPKPpg

2024-07-23 07:38:10 Who’s a good puppy?! ⁦@soumithchintala⁩ #ICML2024 https://t.co/eWjktS7D5T

2024-07-23 07:14:27 RT @jparkerholder: Excited to announce that Genie will be presented as an Oral at @icmlconf #ICML2024, see you all in Vienna!!

2024-07-20 10:44:42 RT @AlphaSignalAI: Microsoft is about to crack the way LLMs understand spreadsheets. Their new SpreadsheetLLM encodes spreadsheet contents…

2024-07-20 10:43:05 RT @_akhaliq: Scaling Retrieval-Based Language Models with a Trillion-Token Datastore Scaling laws with respect to the amount of training…

2024-07-20 10:39:39 RT @conormdurkan: Udio is hiring a Machine Learning Engineer in NYC. You'll work on building, evaluating, and deploying state-of-the-art…

2024-07-20 10:39:11 RT @_rockt: At @ICMLconf 2024 in Vienna next week, Google DeepMind's Open-Endedness Team will be presenting two Orals! "Genie: Generative I…

2024-07-20 10:36:25 RT @maxjaderberg: So many great new papers in diffusion space over the last few weeks, really highlights the flexibility of this modelling…

2024-07-18 20:31:39 RT @OpenAIDevs: Introducing GPT-4o mini! It’s our most intelligent and affordable small model, available today in the API. GPT-4o mini is s…

2024-07-18 19:56:21 RT @brandondamos: In our new @UncertaintyInAI paper, we do neural optimal transport with costs defined by a Lagrangian (e.g., for physica…

2024-07-17 17:02:15 RT @_akhaliq: Video Occupancy Models We introduce a new family of video prediction models designed to support downstream control tasks. We…

2024-07-17 06:22:47 RT @Figure_robot: Figure + BMW Group's Spartanburg Plant → Fully autonomous → AI-driven vision model → Neural Networks for all grasps htt…

2024-07-17 05:53:29 RT @Parskatt: Pretty fun paper, finetuning llama to produce blender code for synthetic renderings https://t.co/qv5quYwar2

2024-07-12 11:21:11 RT @_philschmid: Good data is all you need! How Synthetic Data won AIMO 2024 Progress prize: 1. Used @deepseek_ai math base 7B model 2. Cu…

2024-07-12 11:20:11 RT @PyTorch: Introducing FlashAttention-3 Fast and Accurate Attention with Asynchrony and Low-precision. Thank you to @colfaxintl, @AIat…

2024-07-12 04:41:40 @dpkingma @dileeplearning @icmlconf There’s this @dpkingma : https://t.co/XuJptkCf8l

2024-07-12 04:34:59 RT @caglarml: Recently, there have been many discussions on whether LLMs have a form of self-awareness and recognize their own outputs. In…

2024-07-12 04:28:33 RT @_akhaliq: Autoregressive Speech Synthesis without Vector Quantization We present MELLE, a novel continuous-valued tokens based languag…

2024-07-10 12:04:48 RT @jaseweston: Distilling System 2 into System 1 - System 2 LLMs spend compute to improve responses (CoT, BSM, RaR, Sys 2 Attention, ..…

2024-07-09 22:05:22 @dileeplearning @icmlconf But sadly Californian companies enforce noncompetes in other places. Imagine a preferential AI for California and another for the rest of the world! Misaligned values.

2024-07-09 15:47:20 I should add these are my opinions, but I would love to hear other opinions to understand this better, and revise my beliefs accordingly. There might be room for compromise too. 6-month and 1-year non-competes seem awfully long in a field that moves so fast. Would a max of 3… https://t.co/RATiRQlUZX

2024-07-09 15:03:17 RT @arankomatsuzaki: Google presents On scalable oversight with weak LLMs judging strong LLMs https://t.co/8kKA3MpLom https://t.co/umqmA1N…

2024-07-09 14:48:12 AI scientists and engineers need to make their voices heard when it comes to non-competes. I hope this becomes a topic of discussion at @icmlconf and other venues. We urgently need more activism in this area. I believe non-competes should be banned because: 1. They give… https://t.co/7eARLbldec

2024-07-05 15:24:27 Sir Alex Younger is exceptionally bright and has a deep knowledge of geopolitics. One of the smartest people I’ve ever met. From him and Marc Andreessen the message is clear: The West has to cooperate and *lead* responsibly and decisively in AI to preserve the freedoms and… https://t.co/w00LR8ysCz https://t.co/peREDqlxOx

2024-07-05 15:06:00 RT @Thom_Wolf: The @kyutai_labs fully end-to-end audio model demo of today is a huge deal that many people missed in the room Mostly irre…

2024-07-03 07:39:26 RT @neilzegh: Join us tomorrow to learn about our recent work on multimodality!

2024-07-03 07:39:20 @neilzegh Looking forward to it

2024-07-03 07:37:53 @kroscoo @GoogleDeepMind Keep enjoying your amazing journey, Kris!

2024-07-02 15:32:35 @david_picard I think focusing on a simple concept of information theory is missing the point. The issue here is more practical. If you need detailed captions for 5 billion videos, it would take you a huge amount of effort, incentives, and money to get good data from humans. If you google alt… https://t.co/FVS9ZyPn0q

2024-07-01 23:14:07 @david_picard I took the liberty of asking Copilot, and it answered the following: “Let’s explore the interplay of data processing inequality, rate distortion, information bottlenecks, and representation in the context of using synthetic captions from LLMs to train a text-to-image diffusion… https://t.co/WqhUiYD1vS

2024-06-23 14:44:55 RT @SebastienBubeck: Some of you might find this short interview interesting. Quite a journey with @EldanRonen since the entropic barrier…

2024-06-23 14:42:04 For people interested in the pros and cons of artificial intelligence, I highly recommend reading or listening to this authoritative and superbly thought out book. Admittedly, as an “AI expert” (apologies for the over-used term), I was a little bored at the beginning, but as I… https://t.co/1rKn4yzbOZ https://t.co/3M2ClRTqGe

2024-06-23 13:58:43 This is a wonderful weekend read. Biological intelligence is one of the true miracles of nature. https://t.co/JRbwLJWoeN

2024-06-21 15:28:31 I had VSCode open last night when one of my girls walked past and said "that looks scary". I opened a Jupyter cell with @MSFTCopilot and typed "Please write a function that adds two numbers". It did it and quickly my daughter, who hadn't coded before, started asking it to do… https://t.co/5S7ouEOqit

2024-06-20 08:26:40 RT @lmsysorg: Chatbot Arena update! @NVIDIAAI's Nemotron-4-340B has just edged past Llama-3-70B to become the new best open model on Arena…

2024-03-01 00:00:00 CAFIAC FIX

2024-03-11 00:00:00 CAFIAC FIX

2023-05-22 22:13:52 RT @StanfordHAI: Generative AI begs answers to thorny questions about art authenticity, valuation, compensation, and copyright, and provoke…

2023-05-22 22:07:39 RT @_akhaliq: LIMA: Less Is More for Alignment LIMA, a 65B parameter LLaMa language model fine-tuned with the standard supervised loss on…

2023-05-22 21:53:55 RT @svlevine: We figured out how to train diffusion models with RL to generate images aligned with user goals! Our RL method gets ants to p…

2023-05-20 15:20:58 RT @GoogleAI: Today, we discuss the current state of differentially private ML (DP-ML) research with an overview of common techniques for o…

2023-05-20 11:04:28 @RRejeleene @DeepIndaba @ptrmadurai @mkstalin @CMOTamilnadu @Veera284 @svembu @MahesanNiranjan Do you know of similar efforts to the Indaba in TN?

2023-05-19 19:00:00 CAFIAC FIX

2023-05-21 19:00:00 CAFIAC FIX

2023-05-19 09:39:57 RT @DeepMind: How can we use AI to make the world around us more accessible? Today, we’re proud to help launch a new visual question and…

2023-05-19 09:33:20 RT @pcastr: I met lots of fantastic African researchers while I was in Rwanda. Many of them got started with organisations like Inaba. I w…

2023-05-19 09:32:20 RT @kbeguir: If we come together we will reach our goal, pls donate for @DeepIndaba!

2023-05-19 09:31:22 The @instadeepai team, an AI startup out of Africa, has done wonders to advance AI tools to fight all sorts of disease. Diversity and inclusion in AI is making a very positive and essential difference in our world. This is only the beginning. https://t.co/XvMhvJEUvx

2023-05-19 09:22:40 RT @instadeepai: We're proud to have sponsored this @DeepIndaba Grand Challenge to combat Leishmaniasis

2023-05-19 09:22:20 RT @sgowal: DeepMind is dedicated to bringing Safe, Reliable &

2023-05-19 09:18:12 RT @smhall97: The Deep Learning Indaba has launched its crowdfunding campaign to help support attendees travel to the upcoming Indaba in Ac…

2023-05-19 09:16:16 RT @GoogleAI: Learn how we turned a Vision Transformer image encoder into an efficient video backbone using sparse video tubes (3D grid-bas…

2023-05-19 09:15:02 RT @sundarpichai: More AI-powered accessibility updates, including a feature using a visual language model to describe images without alt t…

2023-05-19 09:14:22 RT @AmalRannen: Would you like to contribute to bringing together hundreds of Machine Learning students, professionals, experts and enthusi…

2023-05-19 09:12:56 RT @sangmichaelxie: Should LMs train on more books, news, or web data? Introducing DoReMi, which optimizes the data mixture with a small…

2023-05-19 09:12:30 RT @mathwis_emily: Hey! did you know that we are crowdfunding for the Deep Learning Indaba this year in Ghana? Check out video + testimoni…

2023-05-19 09:12:11 RT @StanfordHAI: Complex code written by large language models is prone to failure, as a single mistake can break the entire program. Parse…

2023-05-19 09:11:43 @avishkar58 @JeffDean Thank you for your continued support Jeff. It has made a massive difference, and it helped kick start the @DeepIndaba and @Khipu_AI. I’m looking forward to seeing you in Ghana soon.

2023-05-19 09:06:14 RT @JeffDean: The Hopper-Dean Foundation is proud to help support the Deep Learning Indaba. Please consider doing the same!

2023-05-19 09:05:38 RT @boazbaraktcs: Supreme court rules Warhol's transformations of Prince's photos are not "fair use" https://t.co/CVeG0COyCg . We mentioned…

2023-05-19 09:05:12 RT @CohereAI: Generative AI has made great strides, producing images, text, video &

2023-05-19 09:02:04 @DynamicWebPaige Accra, Ghana. You’ll make a big difference there.

2023-05-19 09:00:05 RT @pabbeel: Wonderful chat with @YejinChoinka on instilling AI with common sense and morality! https://t.co/RThNiPcPNh

2023-05-19 08:55:21 RT @avishkar58: We're raising additional money to create opportunities for future African AI leaders to attend the Deep Learning Indaba! Pl…

2023-05-11 08:59:46 RT @Google: Our AlphaFold program accurately predicted the 3D shape of 200M proteins — a breakthrough that gave us the equivalent of 400M y…

2023-05-11 08:59:28 RT @pabbeel: Thank you @geoffreyhinton for diving deeper into the major potential risks you see with AI, and also reminding us of the treme…

2023-05-11 08:51:38 RT @doomie: Fun videos + music created with Phenaki, MusicLM and Bard in the Google I/O pre-show, happening right now! https://t.co/7yzP6c8…

2023-05-11 08:51:03 RT @huggingface: We just released Transformers' boldest feature: Transformers Agents. This removes the barrier of entry to machine learnin…

2023-05-11 08:50:49 RT @NathanLands: AutoGPT? Not useful yet. Code Interpreter, however, is set to revolutionize data science. And it actually works. Here ar…

2023-05-11 08:07:55 RT @demishassabis: As an AI-first company, I'm so excited for what's coming, including Gemini, the new Google DeepMind foundation model in…

2023-05-11 08:07:20 @DimitrisPapail @aminkarbasi @SebastienBubeck Nice

2023-05-10 22:02:13 @aminkarbasi @SebastienBubeck My bard sample I’m also wondering about why this drop in ability after alignment. There’s variance in the samples, but still … https://t.co/TmuJmtnr75

2023-05-10 21:37:09 An incredible opportunity to do a postdoc with Jeff Clune at UBC. I spent 12 years there and loved it. Kindest people. Also Jeff is one of the most creative and brightest minds in AI https://t.co/Zr9z2JTEaZ

2023-05-10 21:31:58 RT @arankomatsuzaki: Large Language Model Programs Presents LLM programs, the emerging methodology of embedding LLMs in a classic program…

2023-05-10 21:31:43 RT @DrJimFan: Finally happening: HuggingFace Transformers Agent. It enables a coding LLM to compose other HF models on the fly to solve mul…

2023-05-10 21:30:38 @JoeFenton Technically speaking, yes

2023-05-10 21:29:21 @doomie At the flying car drive thru

2023-05-10 21:24:04 @patemedom Congratulations, Patrick! It’s wonderful to see your amazing work out

2023-05-10 21:20:48 RT @patemedom: I'm very excited to share the work that comprised my DeepMind internship: “Knowledge Transfer from Teachers to Learners in…

2023-05-10 21:17:16 RT @DeepMind: PaLM-2 is a next generation large language model with improved coding, multilingual and reasoning capabilities. It will powe…

2023-05-10 18:02:53 RT @generatorman_ai: Move over Alpaca, IBM just changed the game for open-source LLMs Dromedary, their instruction-tuned Llama model, b…

2023-05-10 07:47:09 RT @_akhaliq: Principle-Driven Self-Alignment of Language Models from Scratch with Minimal Human Supervision abs: https://t.co/LPvuuxysCr…

2023-05-09 21:53:46 RT @vkrakovna: Great interview with Geoff Hinton on his concerns about humanity losing control of AI: “You have to create subgoals in orde…

2023-05-09 21:45:25 RT @janleike: Really exciting new work on automated interpretability: We ask GPT-4 to explain firing patterns for individual neurons in LL…

2023-05-06 23:05:15 RT @karpathy: Oops haven't tweeted too much recently

2023-05-06 09:30:03 RT @itspetergabriel: Lots of great submissions to the #diffusetogether challenge. Thanks for your hard work so far. We’ve decided to extend…

2023-05-06 09:24:14 RT @korymath: Prompting generative AI models isn’t magic. It’s computer programming. Here’s how to think about writing great prompts. http…

2023-05-06 09:23:15 RT @AlbertQJiang: Baldur: Whole-Proof Generation and Repair with Large Language Models This is such amazing work. Congrats to Emily, Marku…

2023-05-06 09:15:33 @ylecun @Plinz @patalanov We’re in agreement @ylecun. Without intelligent tools, the human race is eventually doomed. So we need to develop intelligent tools, but we need to do so safely and responsibly. Geopolitics and the unpredictable emergence of AI capabilities make it very tricky. Thought is needed.

2023-05-06 08:58:31 @kchonyc I’m sooooo jealous, but so happy to see you there! Also momentous occasion: the first AI conference in Africa

2023-05-05 22:37:11 RT @gabrielpeyre: Slides of my talk of my tutorial talk "the mathematics of neural networks" at the annual meeting of Société Française d'O…

2023-05-05 22:04:51 RT @jeffclune: Seeking a postdoc to join my lab at UBC! Interested in combining deep RL &

2023-05-05 07:30:27 RT @dfgentile: @LangChainAI is a robust framework for building LLM-powered apps, so it is vital to understand the basics My post yesterday…

2023-05-05 07:26:29 RT @MattNiessner: (1/2) NeRSemble #SIGGRAPH'23! We reconstruct dynamic radiance fields for high-quality novel view synthesis of human h…

2023-05-05 07:18:59 RT @DrJimFan: It turns out that human preference can not only be used for RLHF, but also for turning LLM evaluation into a "chess game".…

2023-05-05 07:16:27 RT @ZimingLiu11: To make neural networks as modular as brains, We propose brain-inspired modular training, resulting in modular and interpr…

2023-05-05 07:16:05 RT @arankomatsuzaki: Masked Trajectory Models for Prediction, Representation, and Control Presents Masked Trajectory Models (MTM) as a gen…

2023-05-05 07:11:26 @ilyasut Time to finetune ChatGPT with some Popper philosophy of mind or to add a Theorem Proving plugin It’s not a religion if it allows itself to be falsified.

2023-05-04 21:17:54 RT @IasonGabriel: Personally, I’m not sure much turns on whether the risk posed by AI is “existential”— a term that’s used inconsistently &

2023-05-04 21:17:02 RT @pabbeel: "the old dude that created the AI" (aka the amazing @geoffreyhinton!) will be back on @therobotbrains podcast next week, talk…

2023-05-04 18:24:44 RT @RealRichomie: Just gave GPT-4 access to use Chrome however it wants (click, scroll, fill forms) thru @LangChainAI Auto-GPT agent. The…

2023-05-04 07:50:46 RT @arankomatsuzaki: Visual Chain of Thought: Bridging Logical Gaps with Multimodal Infillings Presents VCoT, a novel method that leverage…

2023-05-04 07:50:12 RT @_akhaliq: Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes reduce both th…

2023-05-04 07:49:59 RT @arankomatsuzaki: CodeGen2: Lessons for Training LLMs on Programming and Natural Languages Releases CodeGen2 models in size 1B, 3.7B, 7…

2023-05-04 07:48:08 @MPI_IS @bschoelkopf @TheOfficialACM Congratulations @bschoelkopf !! Very well deserved.

2023-05-04 00:00:08 RT @_akhaliq: In-Context Learning Unlocked for Diffusion Models Prompt Diffusion. With a prompt consisting of a task-specific example pair…

2023-05-03 23:57:57 RT @aaronlefohn: NVIDIA is thrilled to share our SIGGRAPH 2023 papers. It is our best SIGGRAPH yet, with ~20 papers authored in partnership…

2023-05-03 23:56:17 RT @chipro: New post: RLHF - Reinforcement Learning from Human Feedback Discussing 3 phases of ChatGPT development, where RLHF fits in, ho…

2023-05-03 23:21:21 RT @karpathy: Excellent TED talk from Sal Khan: - many inspiring examples of GPTs finetuned into socratic tutors, assisting without giving…

2023-05-03 23:17:20 RT @arankomatsuzaki: Unlimiformer: Long-Range Transformers with Unlimited Length Input Improves pretrained models such as BART and Longfor…

2023-05-03 09:08:18 @tdietterich @amanpour @NPCollapse @hinton I admire your wisdom Tom, but I must disagree on this. There is nuance. Profs have to market their research to get grants. Geoff did a great job, and it’s thanks to that that we created Deep Learning and trained so many AI scientists. As a Canadian scientist, I am thankful to him

2023-05-03 08:59:03 I share Geoff’s concerns. It is of the utmost need that we continue to develop intelligent tools responsibly, e.g. to fight disease. As scientists and engineers, it is also our moral imperative to point out existential risks when we suspect them. https://t.co/KfrHFMnjHn

2023-05-02 21:29:19 @mustafasuleymn Congratulations!

2023-05-02 21:29:05 RT @mustafasuleymn: Today I’m excited to announce the first version of our new personal AI, Pi... https://t.co/wYpgcXdB1t Pi is smart, kin…

2023-05-01 22:32:54 RT @ziv_ravid: @ylecun and I have been pondering the concept of optimal representation in self-supervised learning, and we're excited to sh…

2023-05-01 22:28:08 I couldn’t agree more. Thanks @geoffreyhinton for your guidance over the years https://t.co/CBc0NcmHtB

2023-05-01 22:21:37 @XingyouSong @GoogleBrain @DeepMind Congratulations !

2023-05-01 22:13:10 @heiga_zen Highly deserved !

2023-05-01 22:08:05 RT @geoffreyhinton: In the NYT today, Cade Metz implies that I left Google so that I could criticize Google. Actually, I left so that I cou…

2023-05-01 08:24:48 RT @_akhaliq: LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model Compared to the original LLaMAAdapter, LLaMA-Adapter V2 can p…

2023-05-01 08:23:44 RT @arankomatsuzaki: Are Emergent Abilities of Large Language Models a Mirage? Presents an alternative explanation for emergent abilities:…

2023-05-01 08:23:40 RT @ylecun: A history of LLMs derived from LLaMA. https://t.co/pz0zcCOBLb

2023-04-30 22:45:28 Where generative AI is going. Movie generation will be great for entertainment, but can anyone think of other uses? https://t.co/QxtqdzyCtx

2023-04-30 22:43:47 RT @ChristianF369: Pushed #gen2 again &

2023-04-29 16:27:15 RT @Grimezsz: I'll split 50% royalties on any successful AI generated song that uses my voice. Same deal as I would with any artist i coll…

2023-04-29 01:17:46 @kooshiar @AndrewYNg @elonmusk @ylecun @geoffreyhinton Interesting hallucination

2023-04-29 01:09:38 RT @pabbeel: Check out our recent work on MTM, a new self-supervised paradigm for RL. MTM aims to reconstruct randomly dropped out elemen…

2023-04-28 17:39:57 RT @forai_ml: Calling ML researchers of Latin America! Join our team and help us ensure that languages from your area of the world are re…

2023-04-28 17:38:14 RT @arankomatsuzaki: Large Language Models are Versatile Decomposers for Table-based Reasoning Surpasses the human performance on Tabfact…

2023-04-28 17:37:18 Prompt to FineTune https://t.co/EhOBxgDJpo

2023-04-28 17:31:15 RT @JoeFenton: This is cool …helping developers graduate from prompting to finetuning. Many more tools like this to come.

2023-04-28 17:30:23 RT @ylecun: A survey of LLMs with a practical guide and evolutionary tree. Number of LLMs from Meta = 7 Number of open source LLMs from Me…

2023-04-28 13:21:23 RT @AndrewYNg: 1/ Thrilled to announce: Our new course ChatGPT Prompt Engineering for Developers, created together with @OpenAI, is availab…

2023-04-28 13:20:30 RT @DeepMind: Football players can tackle, get up, kick and chase a ball in one seamless motion. How could robots master these motor skills…

2023-04-28 07:52:13 RT @pabbeel: New @therobotbrains episode with guest @NandoDF from @DeepMind! We discuss Generalizable AI and how to ensure AI benefits eve…

2023-04-28 07:52:03 Thanks @pabbeel @therobotbrains for giving me a chance to champion @Khipu_AI and @DeepIndaba. I hope more senior AI researchers volunteer in such efforts! Also, great to chat about Google @DeepMind, Gato, AlphaCode, video, future of AI and robotics, and life https://t.co/gqdcb3r7Qx

2023-04-25 06:01:28 RT @_akhaliq: Speed Is All You Need: On-Device Acceleration of Large Diffusion Models via GPU-Aware Optimizations present a series of impl…

2023-04-25 06:00:52 RT @arankomatsuzaki: Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models https://t.co/1kNYQSmJIX h…

2023-04-25 06:00:33 RT @ylecun: An essential step to becoming a scientist is to learn methods and protocols to avoid deluding yourself into believing false thi…

2023-04-24 21:05:42 It’s a brilliant talk by one of the best RL researchers in the world. Recommended. https://t.co/uKOxZIzauf

2023-04-24 05:52:53 RT @arankomatsuzaki: Scaling Transformer to 1M tokens and beyond with RMT By leveraging the Recurrent Memory Transformer architecture, the…

2023-04-23 11:28:21 @yoavgo @johnschulman2 @AdaptiveAgents Again, thanks for helping me satisfy my curiosity, yet your blog touches on broader questions. I love the point about LMs can be trained to be good self-evaluators, replacing the need for human feedback. Could they even be trained to propose self-evaluations?! Learn2learn vibes

2023-04-23 11:14:51 RT @james_y_zou: We should be very cautious when using detectors to classify if text is written by #AI or human. We find these detectors cl…

2023-04-23 11:11:21 @ShamKakade6 @vyasnikhil96 @boazbaraktcs Wonderful paper! Very clear, elucidating and helpful - an important contribution to generative AI! Congrats

2023-04-23 10:48:17 This is a thoughtful take on supervised learning and RL for language models. Good insights and ideas for future exploration and building. https://t.co/JgBJei93YC

2023-04-23 10:44:02 @yoavgo @johnschulman2 @AdaptiveAgents I’m very curious about this, but it maybe that this too is yet another question to be answered empirically. Regardless, I loved you article. It’s clever and thought provoking. The last point on using the LM itself as a general success metric is important. 2/2

2023-04-23 10:31:17 @yoavgo @johnschulman2 That is true, but I wonder if there’s more? In dagger, the LM acts and users (or models) provide feedback, but crucially it is always the LM agent acting. @AdaptiveAgents calls it counterfactual learning to acquire causal knowledge P(O|do(A)) https://t.co/NAFHmPaBTa 1/2

2023-04-22 23:50:26 RT @yoavgo: I was puzzled for a while as to why we need RL for LM training, rather than just using supervised instruct tuning. I now have a…

2023-04-22 23:44:23 @yoavgo @johnschulman2 Roboticists have for a long time considered 3 strategies: behaviour cloning (supervised training), DAGGER, and RL. RL when the reward is known leads to “super-human” performance. Any thoughts on DAGGER in the context of LMs? See e.g. Drew Bagnell’s “an invitation to imitation”

2023-04-22 08:04:32 RT @tyrell_turing: Cool paper showing that during in-context-learning transformers actually recapitulate gradient descent in their forward…

2023-04-21 13:24:08 RT @_akhaliq: #AutoGPT crossed 100,000 Stars on github try out the @Gradio demo on @huggingface to run it easily in the browser demo: h…

2023-04-21 12:25:12 @FelixHill84 Debatable

2023-04-21 12:23:42 @ZoubinGhahrama1 +1

2023-04-21 12:23:18 RT @ZoubinGhahrama1: Truly excited by the merger of Brain and DeepMind to form a new unit: Google DeepMind! Both organizations have a proud…

2023-04-21 12:23:10 +1

2023-04-21 07:03:15 RT @nelsonfliu: Generative search engines are transforming how we find info, but are they trustworthy? We evaluate Bing Chat, NeevaAI, htt…

2023-04-21 00:00:01 CAFIAC FIX

2023-04-21 07:03:15 RT @nelsonfliu: Generative search engines are transforming how we find info, but are they trustworthy? We evaluate Bing Chat, NeevaAI, htt…

2023-04-20 18:08:46 RT @chrmanning: April 2023 AI vibes: Sparks of intelligence flying off GPT-4. A resurgence of open source LLMs. Lots of little Alpacas runn…

2023-04-20 17:59:52 RT @gdb: TED talk from earlier this week. Shows a bit of the future of AI tools, how we teach AIs to follow our intent, and how the tools t…

2023-04-20 17:59:29 RT @jeffclune: I agree. "I'm amazed that people confidently pronounce these things are not sentient, and when you ask them what they mean b…

2023-04-20 17:34:52 RT @dpkingma: Brain and DeepMind merged. Good move for the company imo.

2023-04-20 17:20:58 RT @DeepMind: We’re proud to announce that DeepMind and the Brain team from @Google Research will become a new unit: . Toge…

2023-04-20 17:20:49 I’m super excited about Brain and DeepMind coming together to shape the future of AI. Looking forward to working with old and new colleagues in our Google DeepMind team! https://t.co/1LZtwhQIIc

2023-04-20 17:14:30 RT @demishassabis: The phenomenal teams from Google Research’s Brain and @DeepMind have made many of the seminal research advances that und…

2023-04-16 15:09:27 RT @StefanoErmon: Very excited about this work! A principled way to handle boundaries (e.g., enforce pixel values to be in [0, 255]) in dif…

2023-04-13 05:49:09 RT @JayScambler: The pace of development surrounding Baby AGI and AutoGPT is mind blowing. Seems like a new *groundbreaking* update comes o…

2023-04-13 05:48:46 RT @NathanLands: AutoGPTs are improving at a blazingly fast speed and could soon transform the face of business. Here's what you need to k…

2023-04-13 05:47:54 RT @asimdotshrestha: Introducing #AgentGPT, an attempt at #AutoGPT directly in the browser Give your own AI agent a goal and watch as it…

2023-04-13 05:46:30 RT @SullyOmarr: Whoa.. still not convinced of AI Agents? This might change your mind... I pretended to be a fake shoe company and gave Aut…

2023-04-13 05:45:24 RT @frankc: Agents are game changers I'm building on top of @yoheinakajima 's babyagi code, backed by Pinecone + Slack. Each thread has ne…

2023-04-12 14:48:39 RT @_akhaliq: Zero-Shot Video Editing Using Off-The-Shelf Image Diffusion Models @Gradio demo is out on @huggingface Spaces demo: https://…

2023-04-11 05:36:39 RT @_akhaliq: Harnessing the Spatial-Temporal Attention of Diffusion Models for High-Fidelity Text-to-Image Synthesis abs: https://t.co/GJ…

2023-04-11 05:35:58 RT @arankomatsuzaki: WebBrain: Learning to Generate Factually Correct Articles for Queries by Grounding on Large Web Corpus Constructs a l…

2023-04-10 19:09:16 RT @sbdzdz: Such a cool paper! Jointly optimizing shape parameters, albedo, and roughness from a few reference images through differentiabl…

2023-04-10 19:05:02 RT @michael_nielsen: Interesting list of 137 emergent properties of large language models, by @_jasonwei: https://t.co/hL3P23Gi2h Defined…

2023-04-10 19:00:07 RT @karpathy: Love it - much fertile soil for indie games populated with AutoGPTs, puts "Open World" to shame. Simulates a society with a…

2023-04-10 13:15:08 RT @Teknium1: An even bigger than GPTeacher self instruct dataset from GPT-4 ->

2023-04-10 13:11:01 RT @lmsysorg: We are excited to release the weights of Vicuna-13B. Run it with a single GPU on your own machine! Get the weights: https…

2023-04-08 22:26:11 RT @omarsar0: 8 Things to Know about LLMs If you are working with LLMs, it's worth checking out this paper. It discusses important consi…

2023-04-08 07:25:13 RT @heyBarsee: ChatGPT is just the tip of the iceberg. 2,000+ AI tools were released in March. Here are the 26 AI tools you cannot miss:…

2023-04-07 22:47:38 RT @_akhaliq: Instruction Tuning with GPT-4 abs: https://t.co/sJS32624FU project page: https://t.co/2zo33Z0fy2 github: https://t.co/7QGZ…

2023-04-07 07:11:12 RT @_akhaliq: "The Stormtroopers, on the beach" https://t.co/FjylanWzXk

2023-04-07 07:10:56 RT @karpathy: The analogy between GPTs of today to the CPUs of early days of computing are interesting. GPT is a funny kind of programmable…

2023-04-07 07:10:17 @FelixHill84 There are four kids in the classroom. The teacher brings 30k . How many does each kid get?

2023-04-06 23:32:44 One helpful way to think about self improvement: Optimise E_p(x)pi(y|x) [ Q(x,y) log pi(y|x) ]. The expectation E is wrt the input data x and the model samples y. Q is a surrogate for expected return (what we value, reward, filter). The magic is policy gradients https://t.co/Dok6v8sFwD

2023-04-06 23:13:45 RT @lvwerra: Excited to introduce: StackLlama An end-to-end tutorial for training Llama with RLHF on preference data such as the StackExc…

2023-04-06 23:09:31 RT @_akhaliq: Generative Novel View Synthesis with 3D-Aware Diffusion Models abs: https://t.co/ajKnOJvW2a project page: https://t.co/wPEc…

2023-04-06 23:06:33 RT @DeepMind: What could a world with artificial general intelligence look like? We collaborated with @CSM_News students on a new exhibi…

2023-04-06 23:05:30 RT @sedielem: New blog post about diffusion language models: https://t.co/uMF2BZNCqZ Diffusion models have completely taken over generativ…

2023-04-05 22:28:10 @_akhaliq Happy Easter Enjoy your week off

2023-04-05 22:26:43 RT @TheTuringPost: The second part of the @fastdotai free course is out! "From Deep Learning Foundations to Stable Diffusion" will teach y…

2023-04-05 22:25:54 RT @gdb: A nice note from a ChatGPT user: https://t.co/mlvU3cVTQf

2023-04-05 22:18:10 RT @pulkitology: Introducing DribbleBot: A robot that can dribble a soccer ball on diverse natural terrains. Be it snow, be it grass, be it…

2023-04-05 22:15:53 RT @syhw: Do you need to quantize models? Try diffq, `pip install diffq` and https://t.co/oQLlPHw6Ek

2023-04-05 22:15:33 RT @tejasdkulkarni: It zero-shot discovers all entities in Montezuma's revenge. #AGI

2023-04-05 22:06:52 This was fun. 3/N was hilarious https://t.co/mZPqGodKD2

2023-04-05 19:53:53 RT @pabbeel: New episode, with @aidangomezzz from @CohereAI! We discuss the Transformers paper, Large Language Models, Command and Instruc…

2023-04-05 19:50:24 RT @MetaAI: Today we're releasing the Segment Anything Model (SAM) — a step toward the first foundation model for image segmentation. SAM…

2023-04-05 07:21:18 RT @AlphaSignalAI: A new method of Prompt Injection Attack on GPT-4 was just found! By wrapping the malicious prompt in markdown and ins…

2023-04-05 07:16:58 RT @JeffDean: Paper: TPUv4 system has an optically reconfigurable network to assemble groups of 4x4x4 chips like legos (4x4x12? 16x16x16?).…

2023-04-04 23:18:23 RT @johnjnay: Detailed LLM Evals -Stratified eval can reveal subfields where hallucinations are more likely to occur -LLMMaps: new visual…

2023-04-04 23:06:27 Smaller models with higher quality datasets are becoming quite interesting https://t.co/3TRzUyeYs5

2023-04-04 23:04:18 @tdietterich @compthink I agree Tom. There are also newer modules like S4 that have state, but still experimental. In addition, scratchpads and plugins are often used as external memories even if this is not made explicit. I suspect memory will keep improving when the applications demand it.

2023-04-04 22:52:46 @Khipu_AI Felicitaciones

2023-04-04 22:52:21 RT @DeepMind: Looking for a specific moment in a @YouTube video? Our AI technology will help you get there faster. Over 80 million video…

2023-04-04 08:14:26 RT @alexolshevsky1: I disagree with this and want to explain why. In the thread below, @aryehazan clarifies that his opinion that current…

2023-04-03 22:28:15 Scaling vision transformers to 22 billion parameters. I love the measurement of shape/texture bias in this work. We still need to solve vision! https://t.co/B0Ohy5AeSv

2023-04-03 22:12:05 RT @AndrewLBeam: How will GPT-4 change medicine? Does GPT-4 show empathy? What are the unique risks that GPT-4 poses for healthcare? W…

2023-04-03 19:02:44 RT @yudapearl: Got my first session with GPT-4, amazing! Though it failed its first causal understanding test. Me: "Is it possible that sm…

2023-04-03 18:53:42 RT @AlphaSignalAI: Impressive. LLMs can self-improve without additional training data, reinforcement learning, or human intervention. 1. G…

2023-04-03 05:34:24 RT @arankomatsuzaki: Self-Refine: Iterative Refinement with Self-Feedback Presents a novel approach that allows LLMs to iteratively refine…

2023-04-03 05:28:59 RT @SigGravitas: Massive Update for Auto-GPT: Code Execution! Auto-GPT is now able to write it's own code using #gpt4 and execute pytho…

2023-04-03 05:25:11 RT @sarahookr: we don't like hallucinations when it relates to factually incorrect outputs. however, it is also hallucinations that allow L…

2023-04-03 05:24:32 RT @karpathy: Next frontier of prompt engineering imo: "AutoGPTs" . 1 GPT call is just like 1 instruction on a computer. They can be strung…

2023-04-01 14:03:13 @SIfill_ @timnitGebru @DAIRInstitute Absolutely agree

2023-04-01 10:07:38 RT @AlphaSignalAI: Just came across Vicuna, an open-source chatbot impressing GPT-4. Vicuna-13B achieves >

2023-04-01 10:06:05 RT @DeepMind: We put our AI system AlphaFold in the hands of researchers.  Now, it’s transforming how biology is being done around the w…

2023-04-01 10:05:39 RT @ClementDelangue: All companies will train their own chatgpt/GPT4 thanks to open-source! So cool to see this paper from Bloomberg, whic…

2023-04-01 10:05:04 RT @ronithhh: Generating mini macOS apps with natural language! https://t.co/mAK1EaBqyJ

2023-04-01 02:21:50 RT @hturan: here’s a force-directed knowledge graph interface for @OpenAI’s gpt-4. given a topic, it prompts new questions to ask based on…

2023-04-01 02:08:03 RT @GoogleAI: Learn about ViT-22B, the result of our latest work on scaling vision transformers to create the largest dense vision model. W…

2023-04-01 02:05:17 @iassael @EmtechEurope Huge congratulations Yannis well deserved

2023-03-31 07:47:29 RT @arankomatsuzaki: Token Merging for Fast Stable Diffusion Speeds up image generation by up to 2x and reduce memory consumption by up to…

2023-03-31 07:47:09 RT @rsalakhu: Research Brief: Breakthrough Enables Perfectly Secure Secret Communications https://t.co/5wdFY9ccl9

2023-03-30 23:56:51 RT @fhuszar: Autoregressive Models, OOD Prompts and the Interpolation Regime My notes on how I've started thinking about understanding indu…

2023-03-30 23:56:21 Building a DOS ChatGPT client in 2023 - Why??? This brings back some traumatic lab memories https://t.co/WoUEQGZsPf

2023-03-30 22:58:24 RT @_akhaliq: Improving Code Generation by Training with Natural Language Feedback abs: https://t.co/cGJyR94RHU github: https://t.co/lkQ…

2023-03-30 22:54:50 RT @_akhaliq: HOLODIFFUSION: Training a 3D Diffusion Model using 2D Images abs: https://t.co/981GQT5i8W https://t.co/XgS0PrXvWn

2023-03-30 22:53:27 RT @arankomatsuzaki: TaskMatrixAI: Completing Tasks by Connecting Foundation Models with Millions of APIs https://t.co/D6TxWkwt16 https://…

2023-03-30 22:52:30 RT @instadeepai: 1/ We are happy to announce the open-source release of the inference code and weights of our four genomics #LLM, the nucle…

2023-03-30 22:50:02 @DeepMind @demishassabis Congratulations! Highly deserved

2023-03-30 22:49:45 RT @DeepMind: Congratulations to @DemisHassabis, John Jumper on receiving the 2023 Canada Gairdner International Award on behalf of the Alp…

2023-03-30 22:48:39 RT @fhuszar: @yudapearl This theorem would be relevant if the causal graph we cared about described the causal relationships between consec…

2023-03-30 22:48:14 @fhuszar @yudapearl It’s a beautiful observation @fhuszar

2023-03-30 22:40:56 RT @yudapearl: GPT-4 should know that my statements about DL never rising above "curve fitting" is not an opinion, nor a "common concern" b…

2023-03-30 00:13:54 RT @RishiBommasani: Foundation models are transforming society: in the past month alone, we've seen a flurry of releases! GPT-4, Claude, P…

2023-03-30 00:00:28 Well explained Andrew. Thank you. https://t.co/e3iUkbNL5L

2023-03-29 23:56:49 RT @AndrewYNg: 1/The call for a 6 month moratorium on making AI progress beyond GPT-4 is a terrible idea. I'm seeing many new applications…

2023-03-29 23:54:30 RT @DavidDeutschOxf: Trying the real ChatGPT-4. It's no better on Popper. https://t.co/Cz1ikZ1zci

2023-03-29 23:50:49 RT @pabbeel: New episode, with @alexandr_wang from @scale_AI! We discuss Data, Labeling, Foundation Models, LLMs, truthfulness, RLHF, AI fo…

2023-03-29 23:50:30 RT @MelMitchell1: I didn't sign "the letter". Current AI poses lots of risks, but describing these systems as "ever more powerful digita…

2023-03-29 23:42:39 RT @SebastienBubeck: I personally think that LLM learning is closer to the process of evolution than it is to humans learning within their…

2023-03-29 23:07:06 RT @tryolabs: Looking to explore the world of AI? Check out our recap of @Khipu_AI 2023! Featuring insights from top researchers like @Rub…

2023-03-29 22:36:47 RT @genmoai: Announcing Genmo Chat, a creative copilot that uses GPT-4 and a large suite of generative AI tools to create and then edit any…

2023-03-29 22:35:58 RT @SciTechgovuk: Think AI is scary? It doesn't have to be! The UK’s new approach to regulation will unleash the benefits of AI and create…

2023-03-29 22:35:10 @SebastienBubeck Fully agree. I also think of training an LLM as recreating human and cultural evolution to some extent. It also includes bits of what humans learn in a lifetime. It’s amazing how fast we can do it.

2023-03-29 22:27:10 RT @_akhaliq: Your Diffusion Model is Secretly a Zero-Shot Classifier abs: https://t.co/IRPxpLJYeu project page: https://t.co/niW6ottKXk…

2023-03-29 22:23:39 @FeryalMP I enjoyed our brainstorms too. It’s amazing how it’s happening now and few grasp it’s significance or how hard we thought it would be.

2023-03-29 22:19:46 @Mvandepanne Hi Michiel, I learned to appreciate this view while teaching cognitive systems at UBC. I gave a talk on it at a NeurIPS workshop — I remember Josh Tenenbaum liking it — but it’s only with LLMs that it has become obvious. It’s a powerful idea.

2023-03-29 07:57:15 Externalism provides one of the most important perspectives on the future of LLMs. They use tools like search and scratchpads to store and retrieve. They create and manipulate symbols. They’ll use compilers, people &

2023-03-29 07:47:31 @kareem_carr @svpino @Grady_Booch The model could call the compiler tool, evaluate the code, and use the result for fine tuning of some sort. In regards to abduction, one could use scratchpad prompts like “list a few possible causes of X”. We need a paper measuring induction, deduction and abduction capabilities.

2023-03-29 07:37:47 RT @slava__bobrov: How gene transcription works. DNA to RNA: #biology by D. Berry https://t.co/fnTC5LWsv1

2023-03-29 07:35:11 RT @sirbayes: The AI genie is out of the bottle. Regulation will be very hard. As Tyler says ‘I really don’t think any international cooper…

2023-03-29 07:31:37 RT @AlphaSignalAI: Game changer. You can now run GPT locally on your macbook with GPT4All, a new 7B LLM based on LLaMa. It's completely op…

2023-03-29 07:24:41 RT @AlphaSignalAI: This is big. The Retrieval Plugin allows ChatGPT to have a memory! The model can now remember information from convers…

2023-03-29 07:22:50 RT @_akhaliq: F2-NeRF: Fast Neural Radiance Field Training with Free Camera Trajectories abs: https://t.co/ZijJbWrY55 project page: http…

2023-03-29 07:22:14 @percyliang This is very helpful @percyliang. Could you please share some thoughts about models that generate either images, voices, sounds, music, videos, or motor behaviour. Some of these seem trickier legally even though from an ML perspective they’re the same. Thanks.

2023-03-29 07:05:05 RT @PeterHndrsn: Wondering about the latest copyright issues related to foundation models? Check out the draft of our working paper: Found…

2023-03-29 02:55:50 RT @_akhaliq: Instruct 3D-to-3D: Text Instruction Guided 3D-to-3D conversion abs: https://t.co/QD8iVY2Cq2 project page: https://t.co/ghU…

2023-03-29 02:54:25 RT @arankomatsuzaki: Unmasked Teacher: Towards Training-Efficient Video Foundation Models Using only public sources for pre-training in 6…

2023-03-28 09:33:18 RT @rsalakhu: RLPrompt: Optimizing Discrete Text Prompts with Reinforcement Learning – Machine Learning Blog | ML@CMU | Carnegie Mellon Uni…

2023-03-28 09:33:05 RT @tdietterich: The main short-term risks of LLMs is that they will improve the effectiveness of phishing, cyber attacks, and disinformati…

2023-03-28 09:32:42 RT @arankomatsuzaki: ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks ChatGPT outperforms crowd-workers for several annotation…

2023-03-28 09:31:39 RT @IanOsband: Fantastic talk from @SebastienBubeck on the "Physics of AI": https://t.co/fD1aXCPIaN - Intelligence has emerged: why? how?…

2023-03-27 23:23:31 RT @dwarkesh_sp: Asked @ilyasut (Chief Scientist and cofounder of OpenAI) about - time to AGI - leaks and spies - what's after generative…

2023-03-27 23:20:26 RT @sirbayes: I think Yann is right, but we can fix this by giving corrective feedback. Eg from human or an internal world model. Need clos…

2023-03-27 23:17:28 RT @_akhaliq: Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators @gradio demo now supports pose conditional, ed…

2023-03-27 23:15:52 RT @black_in_ai: The deadline to apply to Black Founders Fund in Africa, Brazil, Europe and US is today!

2023-03-27 23:15:12 RT @dpkingma: Our work on diffusion distillation, led by @chenlin_meng, is being used by @StabilityAI for their upcoming diffusion models

2023-03-27 23:14:46 @karpathy Fully agree. I’m excited to see exploration in this space.

2023-03-27 23:13:43 This is a nice illustration by @ericjang11 of a form of intelligence we’re only beginning to understand — Machines that learn to reliably use language to control their own language generation are getting closer. https://t.co/DkSRN2A1qq

2023-03-27 22:59:35 RT @JayAlammar: Language Models and Machine Learning: What a Time for Language Models https://t.co/GNYe35KuaZ

2023-03-27 22:58:04 RT @_akhaliq: Make-It-3D: High-Fidelity 3D Creation from A Single Image with Diffusion Prior abs: https://t.co/qykLHelrjP project page:…

2023-03-27 22:57:34 If you’d like to talk about AI on live TV, and learn about the fascinating tech entrepreneurship record of @LisaWinning, please DM her. https://t.co/CWMdNsPEkC

2023-03-27 22:37:01 RT @sedielem: This is definitely a problem with AR waveform models, which produce very long sequences (~10^6 steps) and are prone to "going…

2023-03-27 22:35:16 RT @MSFTResearch: What is intelligence? How does it emerge and how do we measure it? Ashley Llorens and machine learning theorist Sébastian…

2023-03-25 18:16:21 RT @SebastienBubeck: At @MSFTResearch we had early access to the marvelous #GPT4 from @OpenAI for our work on @bing. We took this opportuni…

2023-03-25 18:14:01 RT @raphaelmilliere: @ylecun closing his presentation with some conjectures #phildeeplearning https://t.co/K0biNIKY45

2023-03-25 18:09:44 @ylecun has been right eerily often, eg introducing/championing convnets, automatic differentiation, SGD, contrastive learning, the cherry, etc. But, when I presented super-resolution pixel CNNs at CIFAR 2016, he said autoregressive models would be over in 5 years. Maybe https://t.co/V1lVhx9ff3

2023-03-24 06:52:44 RT @karpathy: "How to chat with a 56-page PDF" Good developer-focused YouTube explainer: https://t.co/gNUQ7MhNpp Very excited about the gro…

2023-03-23 23:47:23 RT @arankomatsuzaki: NUWA-XL: Diffusion over Diffusion for eXtremely Long Video Generation - Generates high-quality long videos with both…

2023-03-23 23:44:20 RT @_akhaliq: The Prompt Artists abs: https://t.co/5cXMOR7Uwb https://t.co/vncHFRBVfZ

2023-03-23 23:42:12 RT @karpathy: The vibes when I joined AI in ~2008: - workshops w 50 ppl musing on whether deep learning will ever work - papers w cute toy…

2023-03-22 23:22:58 RT @_ScottCondron: if you're going to call it transformers, it deserves a cool diagram Made using ControlNet on @modal_labs, it's a great…

2023-03-22 23:17:03 RT @_akhaliq: Wavelet Diffusion Models are fast and scalable Image Generators abs: https://t.co/6gSvPKOwB3 github: https://t.co/gDuPGbVA…

2023-03-22 23:14:43 RT @mattdeitke: Introducing Objaverse, a massive open dataset of text-paired 3D objects! Nearly 1 million annotated 3D objects to pave the…

2023-03-22 23:01:29 RT @pabbeel: New @therobotbrains episode is out, with guest @chelseabfinn! We discuss Distribution Shift, Meta-Learning, Editing LLMs, Sin…

2023-03-22 08:07:37 RT @_akhaliq: Text2Room: Extracting Textured 3D Meshes from 2D Text-to-Image Models abs: https://t.co/RJDk9APpQC project page: https://t.…

2023-03-21 23:18:13 RT @simonkalouche: The FIRST commercialized humanoid robot!! Congrats @agilityrobotics @JonathanWHurst https://t.co/h4XlMH3p01

2023-03-21 23:18:03 RT @agilityrobotics: Detecting people and navigating around them is just one example of a human-centric robot designed with safety in mind.…

2023-03-21 23:10:05 RT @forai_ml: We're excited to host @Tim_Dettmers for the latest edition of C4AI Technical Talks! Join us on March 29th for Tim's present…

2023-03-21 23:06:33 RT @MetaAI: The Casual Conversations v2 dataset is a consent-driven dataset of recorded monologues to enable researchers to evaluate fairne…

2023-03-21 23:02:33 RT @amasad: Convinced wee are witnessing the birth of a new kind of computer. From: Memorizing Transformers https://t.co/ou5oXr9lp0 https:…

2023-03-21 22:55:21 @poolio @BenMildenhall @ajayj_ @jon_barron Congratulations

2023-03-21 22:55:12 RT @poolio: Excited to share that DreamFusion has won an Outsanding Paper Award at #ICLR2023: https://t.co/7UguEPweHw Thanks to amazing co…

2023-03-21 20:50:38 @Thom_Wolf @sir_deenicus @aaron_defazio @davidchalmers42 @ilyasut I agree

2023-03-21 20:43:22 RT @__nmca__: Jack Rae has been on about this for ages, see his great recent Stanford lecture: https://t.co/ssBpmdOhpX

2023-03-21 20:39:37 RT @NVIDIAGTC: From computer vision and conversational AI to autonomous machines and healthcare, learn from the best in the industry and st…

2023-03-21 20:38:41 RT @NVIDIAGTC: Announced today at #GTC23, NVIDIA Grace™ CPU paves fast lane to energy-efficient computing for every data center. With mains…

2023-03-21 20:24:31 RT @JeffDean: Bard is now available in the US and UK, w/more countries to come. It’s great to see early @GoogleAI work reflected in it—adva…

2023-03-21 09:36:57 RT @alexgraveley: The speaks on prompt engineering https://t.co/WRFzaKRZLg

2023-03-21 09:36:12 RT @yining_shi: I got beautiful and promising videos after the first few tries with Gen-2! Check it Gen-2: Text to Video: https://t.co/oZoK…

2023-03-21 07:23:31 RT @NandoDF: @sir_deenicus @aaron_defazio @davidchalmers42 @ilyasut In the end, it took many exceptional engineering achievements, some inv…

2023-03-21 07:23:15 RT @NandoDF: @sir_deenicus @aaron_defazio @davidchalmers42 Thanks @sir_deenicus. Markus Hutter, @ilyasut, Matt Mahoney, us, knew that seque…

2023-03-21 07:22:35 RT @sir_deenicus: @aaron_defazio @davidchalmers42 Oh, In 2011, Knoll &

2023-03-21 07:22:33 RT @sir_deenicus: @aaron_defazio @davidchalmers42 This is still the best theoretical starting point to understanding why LLMs work so well…

2023-03-21 07:22:16 RT @sir_deenicus: @aaron_defazio @davidchalmers42 I think it's actually Matt Mahoney and Jim Bowery who discussed it most clearly, back in…

2023-03-21 07:20:32 @sir_deenicus @aaron_defazio @davidchalmers42 @ilyasut In the end, it took many exceptional engineering achievements, some invention, failure, distractions, and perseverance, but we got here as a community - as a large community, not just a few.

2023-03-21 07:17:50 @sir_deenicus @aaron_defazio @davidchalmers42 Thanks @sir_deenicus. Markus Hutter, @ilyasut, Matt Mahoney, us, knew that sequence prediction with multimodal tokens was the way to AI back then. We talked to each other. That influenced GPT, Gato, etc. We didn’t have the compute, the right architecture, ADAM, attention, …

2023-03-21 07:09:04 RT @davidchalmers42: who saw LLMs coming? e.g. decades (or even 5+ years) ago, X said: when machine learning systems have enough compute a…

2023-03-21 07:01:32 RT @_akhaliq: Towards End-to-End Generative Modeling of Long Videos with Memory-Efficient Bidirectional Transformers abs: https://t.co/CHe…

2023-03-21 06:59:49 RT @_akhaliq: CLIP goes 3D: Leveraging Prompt Tuning for Language Grounded 3D Recognition abs: https://t.co/u68F0hPt6G project page: http…

2023-03-21 06:58:31 RT @_akhaliq: Rotating without Seeing: Towards In-hand Dexterity through Touch abs: https://t.co/d0PgYtTTOK project page: https://t.co/KY…

2023-03-21 06:57:06 RT @_akhaliq: Legs as Manipulator: Pushing Quadrupedal Agility Beyond Locomotion abs: https://t.co/Ybi90c8dn1 project page: https://t.co/…

2023-03-21 00:38:20 RT @NVIDIAGTC: Join us tomorrow at 10 a.m. to hear from @demishassabis, CEO and Founder of @DeepMind share how #AI is advancing scientific…

2023-03-21 00:11:07 Interesting pop use cases https://t.co/UzkdXsdMGj

2023-03-20 23:58:26 RT @ClementDelangue: Excited for the new version of the largest open code dataset. More sub-datasets + better opt-outs. What models/feature…

2023-03-20 23:52:21 RT @osanseviero: El Hackathon de @SomosNLP_ regresa - Los LLMs hablan español Con un enfoque en los Objetivos de Desarrollo Sostenible,…

2023-03-20 23:51:29 RT @_akhaliq: Vision-Language Models as Success Detectors abs: https://t.co/phFyWMafwG https://t.co/yfC9c0aZSj

2023-03-20 23:47:37 RT @RLanceMartin: I built an app that uses ChatGPT for question-answering over all 365 episodes of the @lexfridman podcast. Uses @OpenAI Wh…

2023-03-20 19:55:27 @d_yuqing @DeepMind Exceptional internship! Congratulations Yuqing!

2023-03-20 19:41:09 RT @serkancabi: A step towards general purpose success detectors. Vision-language models are opening up new possibilities.

2023-03-20 19:40:30 Our team showed that Vision Language Models, like (tested) GPT-4 or PaLI, can be used no only as policies, but also as success models, which include reward models and evaluators. Learned evaluators could provide a path for scaling automatic feedback tools. https://t.co/Per80tldWl

2023-03-20 19:28:21 RT @d_yuqing: How can we develop more generalisable reward models for agent behaviours? Excited to share my @deepmind internship project,…

2023-03-20 17:47:49 RT @d_yuqing: How can we encourage RL agents to explore human-meaningful behaviors *without* a human in the loop? @OliviaGWatkins2 and I a…

2023-03-20 07:04:08 RT @_akhaliq: CHAMPAGNE: Learning Real-world Conversation from Large-Scale Web Videos abs: https://t.co/ckvhDaTPho project page: https://…

2023-03-20 07:00:10 RT @arankomatsuzaki: CoLT5: Faster Long-Range Transformers with Conditional Computation Achieves: - stronger performance than LongT5 with…

2023-03-19 22:24:38 RT @_akhaliq: 1.7 billion parameter text to video diffusion model @Gradio demo is out on @huggingface demo: https://t.co/o7gsJRhvcx model…

2023-03-19 10:14:57 RT @ScienceMagazine: Last year in Science, researchers presented a novel approach for recycling mixed plastic waste into useful chemicals.…

2023-03-19 10:12:59 RT @Ishwariya13: My dad left us 7 years ago today. He quit his work &

2023-03-18 20:25:58 An AI-centric view: Time appears to be a tool that we use to reason about the world, e.g. make predictions about the world, coordinate with other people, plan, etc. It is harnessed with symbols, calendars, clocks, etc. @lexfridman https://t.co/m5Xx8VzYBO

2023-03-18 20:16:13 RT @JiliJeanlouis: mind blowing ! Stable diffusion running in browser without server https://t.co/rmjBLywh4X

2023-03-18 20:13:41 RT @sirbayes: I finally installed github copilot (better late than never :). Holy cow, it's awesome!

2023-03-18 20:12:31 RT @liuziwei7: #CVPR2023 "Taming Diffusion Models for Audio-Driven Co-Speech Gesture Generation" We propose, *DiffGesture*, the first diff…

2023-03-18 20:10:15 @fadeyifemi You’re absolutely right, Femi. There exist mechanisms to stop deployed LLMs from spouting every generated token. However they seem far from a mechanism that could arrive at “I think therefore I am” - the degree of agency and awareness of a child seems missing for now.

2023-03-17 23:01:34 RT @GoogleAI: Introducing Vid2Seq, a visual language model for dense video captioning that simply predicts all event boundaries and caption…

2023-03-16 22:33:29 RT @_akhaliq: alpaca-lora: Code for reproducing the Stanford Alpaca InstructLLaMA result on consumer hardware github: https://t.co/NB6nrDX…

2023-03-16 22:31:32 RT @DeepMind: We recently released a code update to #AlphaFold 2. The AI system has been trained on new data to produce better results for…

2023-03-16 12:12:22 RT @mathemagic1an: Language modeling seems to be entering it's "Stable Diffusion" phase Dalai: dead simple way to run LLaMa on your comp…

2023-03-16 12:03:49 RT @_akhaliq: Highly Personalized Text Embedding for Image Manipulation by Stable Diffusion abs: https://t.co/Z0Eju5YTZi https://t.co/ilg…

2023-03-16 11:16:27 RT @KristenDiCerbo: Adventures in how @khanacademy turned #GPT4 into a tutor. A thread https://t.co/Otrmch0mz1

2023-03-16 09:51:24 Option 1 (Skinnerian): Control AI output with RLHF Option 2 (Gregorian): Teach AI to use language to self-reflect on its unspoken &

2023-03-16 09:40:27 Moreover, current language models say everything they think. Children don’t. And that makes them interesting. “Between stimulus and response there is a space. In that space is our power to choose our response. In our response lies our growth and our freedom.” Viktor Frankl https://t.co/4fE1JJ9kIe

2023-03-16 09:29:18 RT @ancadianadragan: Check out Andreea's work on aligning the representation used for reward functions with what people internally care abo…

2023-03-16 09:27:36 RT @_akhaliq: An implementation of Designing an Encoder for Fast Personalization of Text-to-Image Models by using @huggingface diffusers …

2023-03-16 09:25:30 RT @ylecun: Very nice article by Craig Smith in IEEE Spectrum about the debate on the power and limitations of LLMs between (among others)…

2023-03-16 09:24:58 @kchonyc I remember reproducing those Some of the higher order models made no sense empirically, and researchers quickly found better ways.

2023-03-16 09:22:01 RT @JeffDean: A bit of a peek into the PaLM LLM APIs we announced yesterday (currently in private preview with a few customers/users to get…

2023-03-16 09:21:33 RT @geoffreyhinton: Reinforcement Learning by Human Feedback is just parenting for a supernaturally precocious child.

2023-03-15 19:07:19 RT @sundarpichai: Excited about PaLM API: an easy and safe way for developers to build on top of our language models, and MakerSuite, a too…

2023-03-15 19:06:13 GPT-4 Creator Ilya Sutskever on AI Hallucinations and AI Democracy https://t.co/PkgCJut3S3

2023-03-15 17:32:05 RT @pabbeel: Super-excited to kick off S3 of @therobotbrains podcast with Yoshua Bengio. We discuss LLMs, Higher-Level Cognition, Causalit…

2023-03-15 01:04:17 RT @jordnb: GPT-4 pretty good at just spitting out a full playable pong game in about 4 seconds. https://t.co/YP9me4MhS9

2023-03-15 00:48:27 RT @khanacademy: Did you hear? Khan Academy is using GPT-4 from @OpenAI to shape the future of learning. Starting today, you can sign up…

2023-03-15 00:47:59 RT @duolingo: AI and education make a good duo. Introducing Duolingo Max. A subscription tier above Super that gives you access to your ow…

2023-03-15 00:39:28 RT @geoffreyhinton: @boazhsan Yes, the analogy is not perfect. Also the same billion people can have their knowledge turned into many diffe…

2023-03-15 00:35:37 RT @MichaelTrazzi: GPT-4 can read and summarize the GPT-3.5 paper... using only pixels https://t.co/HH4VLX335b

2023-03-15 00:29:05 RT @adamdangelo: Today we are launching Poe subscriptions, which will provide paying users with access to bots based on two powerful new la…

2023-03-15 00:23:04 RT @AziziShekoofeh: We are so thrilled to announce Med-PaLM 2, our new SOTA medical LLM! Med-PaLM 2 reaches an accuracy of over 85% on US…

2023-03-15 00:22:01 @jluan Congratulations! Very exciting.

2023-03-15 00:04:56 RT @AlphaSignalAI: GPT4 is capable of turning a picture of a napkin sketch to a fully functioning html/css/javascript website. https://t.co…

2023-03-15 00:03:07 It’s amazing to see this computer vision dream being realised https://t.co/tbr4Ay56Tr

2023-03-14 23:54:17 RT @BeMyEyes: We are thrilled to present Virtual Volunteer™, a digital visual assistant powered by @OpenAI’s GPT-4 language model. Virtual…

2023-03-14 23:53:54 RT @_jasonwei: IMO GPT-4 is a bigger leap than GPT-3 was. - GPT-3 advanced AI from task-specific models to a single prompted model that is…

2023-03-14 23:49:10 RT @geoffreyhinton: Caterpillars extract nutrients which are then converted into butterflies. People have extracted billions of nuggets of…

2023-03-14 23:43:46 RT @osanseviero: This week @Khipu_AI took place, an event that gathers amazing people doing ML in academia and industry in/from LATAM. He…

2023-03-14 23:42:35 RT @osanseviero: Very exciting week! Lots of things going on, and the day is still not over!

2023-03-14 23:21:37 RT @tejasdkulkarni: GPT-4 is extremely powerful for coding -- its next level. I suspect coding workflows will now start escaping the trudge…

2023-03-14 20:34:52 RT @sirbayes: Google releases LLMs integrated into Gmail and Docs. Woohoo! https://t.co/foGuPTH0Lk

2023-03-14 20:33:09 RT @arankomatsuzaki: Transformer-based World Models Are Happy With 100k Interactions Outperforms previous model-free and model-based RL al…

2023-03-14 20:31:49 RT @arankomatsuzaki: High-throughput Generative Inference of Large Language Models with a Single GPU Presents FlexGen, a high-throughput g…

2023-03-14 20:28:49 RT @vivnat: Alan sharing our latest @GoogleHealth @GoogleAI research : our new SOTA medical LLM, Med-PaLM 2 On USMLE MedQA, Med-PaLM 2 re…

2023-03-14 18:15:19 RT @AnthropicAI: After working for the past few moths with key partners like @NotionHQ, @Quora, and @DuckDuckGo, we’ve been able to careful…

2023-03-14 18:13:54 RT @arankomatsuzaki: Erasing Concepts from Diffusion Models Can remove concepts from a diffusion model permanently unlike previous methods…

2023-03-14 18:13:31 GPT-4 is absolutely brilliant. I love the focus on education and multilingual, which could empower many millions of disadvantaged people. Congratulations @gdb @ilyasut @sama @woj_zaremba and everyone at @OpenAI https://t.co/oS2eb6FdMQ

2023-03-14 16:19:34 RT @arankomatsuzaki: Self-planning Code Generation with Large Language Model Proposes a self-planning code generation method with LLM, whi…

2023-03-14 16:16:38 RT @JeffDean: Excited to share a number of features that build on many years of @GoogleAI research and generative AI advances: An API to t…

2023-03-13 22:57:43 RT @TheTuringPost: Running LLMs on consumer GPUs. Notes, tutorials, and examples in one thread

2023-03-13 22:52:32 RT @BachFrancis: In this month's blog post, Jensen's inequality, the inequality you often wish is in the other direction. https://t.co/NSAl…

2023-03-13 22:51:42 RT @CIFAR_News: Do differences in the brain predispose some people to loneliness, or does loneliness change the brain? @danilobzdok (@mcgil…

2023-03-13 22:11:55 A bit of cute AI history: @Azure was very supportive of my deep learning research at @CompSciOxford but back in 2014 they had no GPUs. They adapted quickly! @NVIDIAAI, @GoogleAI, Samsung and CIFAR also supported — I’m thankful to them. https://t.co/QjKLFN0Lja https://t.co/g8WhPqh6Wq

2023-03-13 21:13:17 “I’m not saying 12 white men would have avoided this mess, but the company may have been distracted by diversity demands.” According to @WSJ on SVB. Speculation of this kind is nothing but fuel to power our efforts for a more diverse and inclusive world. https://t.co/nDTOp6wkiw

2023-03-13 20:29:09 RT @BlackHC: How good is ChatGPT at Jeopardy!? IBM Watson was a huge effort. Can vanilla ChatGPT compete? Using @LangChainAI and @huggingf…

2023-03-13 15:47:30 @JeffDean @Khipu_AI We missed you

2023-03-13 15:47:13 @tdietterich @JeffDean @Khipu_AI Very happy to hear this Tom!

2023-03-13 06:09:48 RT @_akhaliq: 3D Cinemagraphy from a Single Image abs: https://t.co/DashwTz7ll project page: https://t.co/4ggrn3yrm8 https://t.co/DjxJZQ8…

2023-03-13 06:05:47 RT @karpathy: Dropout layers in a Transformer leak the phase bit (train/eval) - small example. So an LLM may be able to determine if it is…

2023-03-12 17:33:19 RT @Neeva: Considering using a LLM for production, but it's too slow? Well fear not! Today you can with Double Pseudo Labeling! Here's wh…

2023-03-12 17:30:28 RT @OmarUFlorez: Here are some slides I made about how to encode numeric information and combine khipus: https://t.co/zFeYmC3Rdu

2023-03-12 17:26:42 RT @OmarUFlorez: Would love to see how people use dot products and compositionality after combining 2+ khipus as dense layers @NandoDF. Gre…

2023-03-12 17:12:45 RT @_akhaliq: MobileBrick: Building LEGO for 3D Reconstruction on Mobile Devices abs: https://t.co/RL63SU0oZ9 project page: https://t.co/…

2023-03-12 17:08:39 RT @sarahookr: We need more nuanced discussions around the risk of open sourcing models. Open source brings valuable access, but it is ab…

2023-03-12 15:42:10 RT @WindingMichael: We’ve generated a synapse-resolution map of an insect brain, includes 3016 neurons and half a million synaptic sites! O…

2023-03-12 15:39:21 RT @ai__pub: // Toolformer Podcast: Preview // Today I'm interviewing the Toolformer authors! LLMs like Bing (and soon, ChatGPT) can use…

2023-03-12 15:36:15 RT @IasonGabriel: Given the salience of RLHF we need to remember preferences are diffuse, malleable, affected by bias and often constructed…

2023-03-12 15:34:45 Make sure you ask your managers to be supportive. It’s been really tough to raise financial support for these diversity and inclusion efforts, which I feel are essential to attain safe and ethical AI tools.

2023-03-12 15:32:26 It is important for senior AI people to attend AI outreach meetings like @Khipu_AI and @DeepIndaba - Google, DeepMind and Apple have been very supportive. I hope OpenAI, Meta, Amazon, Microsoft etc increase their support. We need to ensure AI will benefit everyone with acts. https://t.co/qWt1bWP7cF

2023-03-12 15:17:14 RT @giffmana: Fine-tuning EfficientNet is the most popular approach in Kaggle, but ViT is catching up. I wonder if FlexiViT, which open-so…

2023-03-12 15:16:20 RT @gmonce: Fue un enorme orgullo para la @montevideoIM y para el @TeatroSolis hacer de anfitriones de este evento público de cierre de #k…

2023-03-12 15:05:46 This is a Khipu layer from Peru, part of a deep Khipu, an ancient computer. Gracias ⁦@OmarUFlorez⁩ — Exercise: Can someone build a transformer core with it? ⁦@Khipu_AI⁩ ⁦@_LXAI⁩ https://t.co/BJ3N1LVUeC

2023-03-12 12:20:30 @pablosprechmann is one of the people who tirelessly works to create AI communities in Latin America - es un campeón! Gracias Pablo! @Khipu_AI @_LXAI https://t.co/Gd3kYtEGE7

2023-03-12 12:15:33 RT @pcastr: My friend @pablosprechmann , one of the main organizers of @Khipu_AI #khipu2023 , giving some closing words. ¡Excelente trabajo…

2023-03-12 12:10:02 @ahirtonlopes @Khipu_AI @sandraavilabr @milalaranjeira @ninadhora Pode ser interessante para você https://t.co/qD2jbTpT3p

2023-03-12 12:04:03 RT @ahirtonlopes: From yesterday, last day of @Khipu_AI with brazilian students and professionals, mostly from Unicamp and USP. Glad to se…

2023-03-12 12:01:45 RT @vpeterson09: Gracias #khipu2023. He tenido la suerte de ir a varias conferencias, ninguna como lo que es @Khipu_AI realmente, lo sien…

2023-03-11 13:58:31 @uniqueneeraj @Khipu_AI https://t.co/U8YAnLMmZO

2023-03-11 13:39:42 RT @omarsar0: Microsoft's Visual ChatGPT is awesome! It connects ChatGPT and different visual foundation models to enable users to interac…

2023-03-11 13:39:10 @jocelyndunstane Samy is definitely one of the nicest guys, and a wonderful champion of education and inclusion. I think he’s a great role model for other guys. Gracias

2023-03-11 13:35:19 RT @ivanafeld: Una banda de rock que llena estadios, @ntvgoficial tocando en #khipu2023 apoyando a las mujeres en IA y a la investigación e…

2023-03-11 13:34:32 RT @jocelyndunstane: I loved #khipu2023 !!! Some photos of today: With Samy Bengio (such a nice guy!), the Chilean team, the view of Teatro…

2023-03-11 13:31:40 RT @PaulHernandez_: The #khipu2023 event has finished. I had the oportunity to attend great talks from Latinamerica speakers working on AI.…

2023-03-11 13:21:43 Diffusion as a neural net, a language model in jax, attention and transformers — some slides from my ⁦@Khipu_AI⁩ tutorial https://t.co/gJ6bZYemot

2023-03-11 12:18:44 RT @IasonGabriel: In the context of AI ethics I usually try to stay positive, but this is deeply upsetting How can anyone be deploying t…

2023-03-11 12:12:15 You made a lot of people very happy. What you did was the greatest thing I’ve seen any music band doing for diversity and inclusion in AI - Gracias @ntvgoficial @Khipu_AI @_LXAI @1benm https://t.co/qviBtDwCnc https://t.co/uEe1M51ilz

2023-03-11 10:59:24 RT @ntvgoficial: Ayer dimos un show en @Khipu_AI: encuentro Latinoamericano de Inteligencia Artificial, organizado por la Facultad de Ingen…

2023-03-11 10:55:00 RT @SebastienBubeck: The Chomsky et al. opinion piece in the @nytimes about ChatGPT is making the rounds. Rather than trying to deconstruct…

2023-03-11 10:54:34 RT @kardaver2: Full livecoding audiovisual set w Pablo Riera at @Khipu_AI #khipu23 ! https://t.co/QApbWD9T8Y

2023-03-11 10:54:04 RT @anitakirkovska: Microsoft launched Visual ChatGPT An agent where you can send/edit images via chat. link: https://t.co/fpaWdmLh4i…

2023-03-11 05:33:32 RT @mfu3ntes: Very inspiring week in @Khipu_AI! Fantastic organization, including talks, panels and even concerts. Here @kchonyc experienc…

2023-03-11 05:31:43 @FortunatoMeire @sandraavilabr @FortunatoMeire: We missed you a lot this year at @Khipu_AI

2023-03-11 05:28:59 RT @SergeiIakhnin: MD Simulation is key to computational drug discovery. At @IsomorphicLabs we are building a combination of physics-based…

2023-03-11 05:17:55 RT @vukosi: Lelapa AICo-founder of Lelapa AI, Pelonomi Moiloa, unpacks how Lelapa AI helps to break the biases that AI has towards Africans…

2023-03-11 05:16:39 RT @pcastr: My dear friend @sarahookr reminds us of the disparate access to opportunities due to geography, the importance of persistence,…

2023-03-11 05:15:16 RT @Khipu_AI: We're hearing from Fabrizio Scrollini, Jocelyn Dunstan, Peter Norvig, Sebastian Barrios, hosted by Luciana Bennoti for the fi…

2023-03-11 05:15:04 RT @_akhaliq: Scaling up GANs for Text-to-Image Synthesis present our 1B-parameter GigaGAN, achieving lower FID than Stable Diffusion v1.5…

2023-03-11 05:14:18 RT @AnthropicAI: Safety is the core research focus of Anthropic and so we’ve written up a post laying out our high-level views on AI safety…

2023-03-11 05:14:03 RT @draix: We’ve shared an unforgettable evening at @Khipu_AI Women in AI event! Besides great panels and special mentions towards women…

2023-03-11 05:13:45 RT @kchonyc: it was perhaps the most majestic venue I've ever given a talk at @TeatroSolis thanks for the opportunity @Khipu_AI https://t.c…

2023-03-11 05:13:26 RT @Khipu_AI: It is the final day of Khipu! This afternoon we are at the beautfiul Teatro Solis to hear AI stories from Latin America and…

2023-03-11 05:12:01 RT @ahirtonlopes: Of course I would take a picture with @NandoDF , leading scientist in Google’s Deep Mind ML team. Thanks for the amazing…

2023-03-11 05:11:28 RT @draix: Today we’ve enjoyed an amazing talk of @NandoDF about Large Language-Vision Models at @Khipu_AI. Nando is @DeepMind Research D…

2023-03-11 05:11:05 RT @lbugnon: Buenísimo el resumen de transformers por @NandoDF (atención a la firmada) https://t.co/PYgM10qB2M

2023-03-11 05:08:45 RT @gustavoq141: Thank you very much for the photo to @NandoDF researcher director @DeepMind with great CONICET researchers @ignaciorlando…

2023-03-11 05:06:57 @caglarml @kchonyc @ArnaudDoucet1 I will miss you dearly @caglarml —you’re an exceptionally bright researcher and one of the nicest people I know — I wish you a very successful next step in your AI career and let’s stay in touch

2023-03-11 04:56:56 Peter Norvig shared a wise perspective on generative models at ⁦@Khipu_AI⁩ https://t.co/mGdEJd461n

2023-03-11 04:54:25 2023 could be a special year for AI music https://t.co/sQi3wKNbho

2023-03-11 04:52:36 @kchonyc also presented thought-provoking questions of governance and influence https://t.co/PqRcuB9vXa

2023-03-11 04:48:43 Brilliant talk by ⁦@kchonyc⁩ at ⁦@Khipu_AI⁩ illustrating the power of generative AI —using a chatbot to generate most of the slides’ text. https://t.co/J82iz8prkA

2023-03-11 04:43:42 RT @kchonyc: Peter Norvig at @Khipu_AI https://t.co/dUPeNh8O6S

2023-03-11 04:43:33 RT @maxjaderberg: It's been well over a year now since joining @IsomorphicLabs here in London, and kicking off ML research. We've been head…

2023-03-11 04:42:39 RT @pcastr: During his great talk, @NandoDF gives a nice shoutout to @RubenEVillegas , an Ecuadorian at Google brain, as an example of how…

2023-03-09 15:33:19 RT @MishaLaskin: New post - how do we train models that are larger than the memory of a single GPU? Break the model into smaller pieces acr…

2023-03-09 12:01:29 My lecture on transformers, VQ-VAE and diffusion models ⁦@Khipu_AI⁩ - I wish I could just append it to my Oxford deep learning course - now with Jax Haiku https://t.co/FoyQchhZYI

2023-03-09 11:55:50 RT @mengjiao_yang: Review paper on Foundation Models for Decision Making: https://t.co/kjkzXIEp5E Foundation models can characterize vario…

2023-03-09 11:55:21 RT @_akhaliq: Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models build a system called Visual ChatGPT, incorporat…

2023-03-09 11:54:26 Proud that @GoogleAI and @DeepMind continue to sponsor education &

2023-03-09 11:32:59 RT @IDATHAuy: Another exciting day full of learning and networking with the #AIlatam Community at @Khipu_AI. If you want to watch our…

2023-03-09 11:27:38 RT @sirbayes: Interesting article from Chomsky. Ignoring "Universal grammar" etc, I think his main point is that LLMs are not learning pars…

2023-03-09 11:26:18 RT @johnjnay: ChatGPT for Training Data 1 ChatGPT rephrases each training sentence into multiple conceptually similar but semantically dif…

2023-03-09 09:02:27 @draix @Khipu_AI @DeepMind @BrainLogicAI Thank you

2023-03-08 20:36:41 I just learned about this startup, https://t.co/AR6yQPMDCR, at ⁦@Khipu_AI⁩ - amazing machine learning demos and technical depth. Idatha and https://t.co/ATVStPzDPE were also very impressive. https://t.co/wJhANa29F9

2023-03-07 22:34:03 RT @MichaelPoli6: Attention is great. Are there other operators that scale? Excited to share our work on Hyena, an alternative to attn tha…

2023-03-06 23:48:49 RT @johnjnay: A Cascade of Foundation Models -Combine diverse prior knowledge DINO vision-contrastive info CLIP lang-contrastive info DALL…

2023-03-06 23:46:13 RT @haoliuhl: Humans learn from rich feedback in the form of language. Why not turning all feedback into a sentence to train models? We pr…

2023-03-06 23:45:30 RT @AnthropicAI: Language models (LMs) exhibit harmful biases that can get worse with size. Reinforcement learning from human feedback (RLH…

2023-03-06 23:43:34 RT @_akhaliq: Unleashing Text-to-Image Diffusion Models for Visual Perception abs: https://t.co/n7jP0p1mDN project page: https://t.co/BBr…

2023-03-06 23:39:56 RT @SanhEstPasMoi: We are reproducing Flamingo, a vision and language model developed by Deepmind (https://t.co/GeLI64VN71). We spent a go…

2023-03-06 23:34:55 RT @BlackInRobotics: We are excited to launch the Black in Robotics &

2023-03-06 23:32:39 @DavidDeutschOxf @sama This is a real practical problem because there exist many undeciphered ancient human languages https://t.co/8hOFnwEXT6 @DavidDeutschOxf @iassael

2023-03-06 23:20:02 RT @Khipu_AI: 'The Khipu' is back at Khipu in Montevideo #khipu2023 https://t.co/sNj3oeIfTf

2023-03-06 23:19:39 RT @DeepMind: They help your muscles move. They carry oxygen in your blood. They let your eyes detect light. Proteins power every process…

2023-03-06 23:18:04 RT @ZoubinGhahrama1: Today we're announcing the Universal Speech Model @GoogleAI as a step in our ambitious commitment to support the worl…

2023-03-05 23:10:23 RT @KatieTarasov: Ahead of #GTC23, I sat down w @nvidia CEO Jensen Huang to hear how his big bet on AI is finally paying off + how it's mit…

2023-03-05 10:00:00 CAFIAC FIX

2023-02-27 21:47:29 RT @arankomatsuzaki: Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback LLM-AU…

2023-02-27 01:00:00 CAFIAC FIX

2023-02-25 22:14:04 RT @johnjnay: Active Prompting for LLMs -Most Chain-of-Thought examples are pulled from a fixed set -Instead, to adapt to diff tasks 1) Fi…

2023-02-24 20:56:46 @ylecun Llama está bien en llamas ! https://t.co/hpyMyq1VIz

2023-02-24 20:46:42 RT @DrJimFan: In 5 years, I believe humanoid robot hardware will finally cross the uncanny valley - so lifelike that you cannot tell if it’…

2023-02-24 08:08:57 RT @yukez: Our new algorithm teaches robots to perform tasks by watching humans play. Check it out!

2023-02-24 08:07:26 @jeffclune @DeepMind Excited to spend time together doing research!

2023-02-23 10:03:57 Thanks @danfei_xu! I’m often amused when I hear: hey have you considered RLHF for robotics?! https://t.co/fB8cTheuNk

2023-02-21 23:38:38 RT @JeffDean: Excited to share the first of a series of @GoogleAI blog posts summarizing our research work from 2022. This covers language…

2023-02-21 23:34:38 RT @hausman_k: We see many examples of LLMs communicating with each other in English to arrive at a conclusion (e.g. https://t.co/ByDDCp8p2…

2023-02-21 23:33:45 RT @tomekkorbak: You can (and should) do RL from human feedback during pretraining itself! In our new paper, we show how training w/ human…

2023-02-21 23:27:44 RT @Khipu_AI: We have some big news... Khipu's big finale will be taking place in the beautiful Teatro Solis and we are opening the doors t…

2023-02-21 07:03:26 RT @arankomatsuzaki: Scaling Laws for Multilingual Neural Machine Translation Examines how increases in the model size affect the model pe…

2023-02-21 07:03:09 RT @McAllesterDavid: https://t.co/ByE6fTE64o A paper on the mathematics of diffusion models that explains the diffusion SDEs --- both forwa…

2023-02-19 10:23:06 RT @johnjnay: Learning to Verify LLM Language-to-Code Gen Train verifiers to predict whether program from CodeLLM is correct based on: -na…

2023-02-19 10:22:30 RT @Jeande_d: Augmented Language Models: a Survey A fantastic survey paper that dives deep into kind of language models that uses reasonin…

2023-02-19 09:30:03 RT @caglarml: I found this fascinating because in my early days at DeepMind, I spent quite a bit of time to build efficient memory models…

2023-02-17 22:31:36 RT @__kolesnikov__: Vision meets RL! We reveal that policy gradient can be used for tuning vision models to optimize complex metrics, such…

2023-02-17 22:30:05 RT @arankomatsuzaki: Text-driven Visual Synthesis with Latent Diffusion Prior Presents a generic approach using latent diffusion modelsas…

2023-02-17 19:25:50 @LelapaAI @kharijohnson @WIRED Hello @LelapaAI ! Sounds exciting

2023-02-17 09:58:59 Happy to see researchers questioning whether RLHF is sufficient for “alignment”. I like this perspective: https://t.co/sW1sHeABBv of @danieldennett. But ethical questions abound as it involves ego, TOM, morality, compassion, empathy… https://t.co/63UF1OgUJL

2023-02-17 09:20:13 RT @Vikashplus: Lack of scale &

2023-02-17 09:16:39 RT @david_sontag: Github Copilot is the most exciting large-scale, rapid, experiment of human-AI interaction that I've ever seen. We're goi…

2023-02-17 09:13:29 RT @DeepIndaba: Are you looking to use Machine learning in your humanities research? Apply to the Deep Learning Indaba to be held in Ghana…

2023-02-16 08:43:42 RT @heiga_zen: A Text-to-Music paper from my colleagues at Google. Noise2Music: Text-conditioned Music Generation with Diffusion Models I…

2023-02-15 23:29:54 @jimmybajimmyba @UofT Highly deserved! Congratulations

2023-02-15 08:08:29 RT @hausman_k: If you want to understand why robotics is much harder than it seems, @ericjang11 pointed me once to this essay that does a p…

2023-02-14 09:40:11 RT @haoliuhl: We introduce an unsupervised method to align text and image. Language Quantized AutoEncoders (LQAE) enables few-shot image c…

2023-02-14 09:29:42 RT @mathemagic1an: My thoughts on Toolformer IMO the most important paper in the past few weeks. https://t.co/4IDciigbkc Teach an LLM to…

2023-02-14 09:13:19 David Guetta says the future of music is in AI. In the creative industries, where imagination is needed, generative models are likely to be very impactful. Many expected creativity and imagination to be one of the last AI achievements, but maybe not so. https://t.co/2TYynlUWtv

2023-02-13 23:36:31 RT @shiraeis: Google research released a paper on a neural net that can forecast rain up to 12 hours ahead, as opposed to using time and co…

2023-02-13 23:26:11 This is a fantastic scaling effort to solve computer vision tasks. Better vision models will help us engineer better safe self-driving cars, and other types of robots https://t.co/451N52P2pQ

2023-02-13 23:11:09 I love the modelling simplicity in this paper: combine 3 transformers into a big transformer and, voilà! amazing results for mapping images+text to text. https://t.co/Rcbc12N9bX

2023-02-13 08:18:35 @drjwrae Jack, can you share the details of your talk with folks here. I know you have a brilliant and inspiring take on this topic. Thanks

2023-02-13 08:13:48 RT @johnjnay: LLM as Agent -LLM grounded in interactive text world w/ online RL (PPO) -Incrementally updating its knowledge w/ observation…

2023-02-13 08:12:18 RT @_akhaliq: Scaling Vision Transformers to 22 Billion Parameters presented ViT-22B, the currently largest vision transformer model at 22…

2023-02-12 19:49:43 RT @johnjnay: "Moral Chain-of-Thought" (MORALCoT) Prompting -Combines LLMs + cognitive sci theories of moral reasoning to predict human ju…

2023-02-12 19:47:52 RT @caglarml: In 1992, I lost my aunt to an earthquake. She died as a hero when she was trying to save a baby from a house. It was the firs…

2023-02-12 18:28:02 RT @AndrewLampinen: Ted Chiang is a great writer, but this is not a great take and I'm disappointed to see it getting heavily praised. It's…

2023-02-12 12:19:38 This is nice thread on LLMs and chatbots by @fchollet. This is as much about understanding LLMs as it is about understanding different aspects of what we think of as human intelligence https://t.co/alfmC50ahu

2023-02-12 08:54:39 RT @JoINrbs: this is such an incredible illustration. stockfish (white) plays chatgpt (black) (source: https://t.co/x66JSrrkTV) https://t.c…

2023-02-12 00:19:51 @francoisfleuret I haven’t gone back to the paper, but looking at this equation alone: When the heads are independent, W_o introduces correlations among them, allowing for information sharing and a denser representation.

2023-02-11 23:08:44 RT @KevinAFischer: This paper is not receiving enough attention: GPT 3.5 displays emergent theory of mind https://t.co/wOcen3ZVTw https:/…

2023-02-11 23:07:44 RT @DimitrisPapail: Can transformers follow instructions? We explore this in: "Looped Transformers as Programmable Computers" https://t.co…

2023-02-11 21:59:51 RT @belindazli: New Paper Language models possess a great deal of common-sense knowledge about real-world environments. How can we take…

2023-02-11 21:59:33 RT @younesbelkada: You asked for it. You can now fine-tune a model that has been loaded in 8-bit. With 8-bit fine-tuning each 1B parameters…

2023-02-11 18:54:14 @sirbayes Don’t upset Kermit the frog

2023-02-11 18:37:06 @BayesianUpdater @sirbayes The expectation is replaced by an average over tokens (a few trillion for the largest LMs) so F can be very general. If a human is selecting the y’s among other y’s to create a dataset, then the human is F, hopefully being sensible

2023-02-11 18:29:47 Finally, when learning from preferences, one learns an F(x,y) that enables one to rank and select or do policy gradients (e.g. PPO) as in most RLHF. When the interface allows for corrections (e.g. rewriting the response in a chat agent), then we are in the domain of Dagger.

2023-02-11 18:27:18 Imitation with Dagger: In counterfactual learning F is typically the identity. The agent acting with policy p(y|x) determines the x’s as in RL, but humans (or other agents) provide corrections in the form of y’s. The new data is used for retraining.

2023-02-11 18:25:19 Self-training: F(x,y) is a filtering/ranking function, eg., what we call a reward/return. The input x may be chosen by humans, but the model generates the y’s and F ranks and selects for further rounds of self-training. F can be explicit or implicit (human in the loop as in RLHF)

2023-02-11 18:18:14 Policy gradients: F = Q(x,y) (the state-action value function), and x and y are generated by the model acting on an environment with policy p(y|x). The x’s are from the invariant state distribution as in the policy gradients theorem.

2023-02-11 18:15:30 Supervised learning: F = I (identity), and x and y are produced by humans. E.g. x is images taken by humans and y are corresponding labels. E.g. 2, x is text and y is the next text token.

2023-02-11 18:13:31 Funny @sirbayes Learning methods — supervised, RLHF, policy gradients, Dagger, self-training — can be seen as optimisation with the following gradient: grad = Expectation_x,y [ F(x,y) grad log p(y|x) ] Choices of F and how x and y are produced determine the learning type 1/n https://t.co/V1uJwO76LW

2023-02-10 10:56:16 Chain-of-thought can also be interpreted as a simple external memory tool: The model writes to a scratchpad, then reads it, and answers the question. The scratchpads to which the model outputs text and inputs text are powerful tools that need further exploration.

2023-02-10 10:54:28 Tools in this context can mean many things. It could be APIs of software products, external memory, and even self-driving cars! It could even be other agents, eg we rely on people to remind us of facts, or to teach us about a new topic.

2023-02-10 10:49:59 Language models (LMs) using tools is one of the most exciting research &

2023-02-09 23:22:39 RT @ancadianadragan: Learning from prefs and demos is more popular than ever, but we have to be careful about the rationality level we assu…

2023-02-09 23:18:25 @erich_elsen Enjoying Brownian motion?

2023-02-09 08:12:49 RT @_akhaliq: Hard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery abs: https://t.co/q2mCZsCe4g…

2023-02-09 08:02:34 RT @sedielem: More text-to-music with impressive results, this time based on diffusion models, rather than autoregression. I suppose any f…

2023-02-09 00:05:45 @ryan_p_adams @rodneyabrooks One of my favourite papers!

2023-02-08 23:49:47 @sirbayes Nicely put, Kevin.

2023-02-08 23:49:20 RT @sirbayes: Great article. Language is not a direct reflection of reality, it is confounded by speaker intent.

2023-02-08 23:46:16 RT @ashkamath20: Check out our new fine-grained vision and language understanding task (CPD) and associated benchmark - TRICD! Contextual…

2023-02-08 21:10:08 RT @AlphaSignalAI: Large Language Models are fascinating. 1. They can store and simulate other NNs inside their hidden layers and adapt to…

2023-02-08 20:21:16 I’m looking forward to attending @Khipu_AI - I hope all speakers do. When students at @DeepIndaba and @Khipu_AI interact with senior researchers we all learn, and it does wonders for their aspirations. https://t.co/8tQpGIY5rL

2023-02-08 20:15:22 RT @yukez: Roomba builds a static map of your home by moving around. Can a robot create articulated models of indoor scenes through its phy…

2023-02-08 09:26:23 While DAGGER is a great idea to enable Feedback for LLMs (eg chat) it is not a replacement for RL because RL opens up room for different forms of feedback (eg preferences). However, as a teacher I would advise careful measurement of the contribution of each to the final metric.

2023-02-08 09:18:23 DAGGER is a form of counterfactual teaching as explained in https://t.co/NAFHmPaBTa - Note that it is the student who always acts. The teacher only provides corrections, which are used to minimise the LLM loss directly. Note however that this imitation IS NOT supervised learning.

2023-02-08 09:14:11 People are asking if there are alternatives to RL in RLHF. Yes, imitation with DAGGER (tutorial: https://t.co/aEMx5xDIOU ). The user provides feedback with corrections, e.g. when the agent says “that” the user tells the agent that instead of saying “that”, it should say “this”. https://t.co/2F9Q2oUXkr https://t.co/8G13t8fEgp

2023-02-07 07:10:50 RT @runwayml: Today, Generative AI takes its next big step forward. Introducing Gen-1: a new AI model that uses language and images to gen…

2023-02-06 23:13:44 RT @ai__pub: Reading the H3 paper this afternoon! Interviewing the authors tomorrow for episode #2 of Deep Papers. Paper: https://t.co/LQ…

2023-02-06 23:01:23 RT @yukez: Nice work on ChatGPT-enabled planning for open-world agents in MineDojo!

2023-02-06 23:00:59 RT @ZoubinGhahrama1: Excited to start introducing Bard, a new LaMDA powered conversational AI system, as well as new AI features in Search!…

2023-02-06 07:05:26 RT @RazRazcle: Is LLM finetuning worth it? If you know what you're doing, finetuned models can be 30x smallerwithout losing performance.…

2023-02-05 22:58:32 RT @rasbt: Tired: train a large language model (LLM) to generate text in human languages Wired: train an LLM to generate proteins sequence…

2023-02-05 22:51:59 RT @bleedingedgeai: Google announces Dreamix: a model that generates videos when given: - video + prompt (Video editing) - input images + p…

2023-02-05 22:50:59 RT @_akhaliq: Tune a video library Browse through Tune a Video models conceptualized and fine-tuned by Community @Gradio demo: https://t.…

2023-02-05 22:50:28 Fascinating work combining chain-of-thought prompting and tool use to improve the reliability of language models. https://t.co/eh2s4mYX7l

2023-02-05 01:53:14 RT @mengjiao_yang: Text-conditioned video generation can serve as universal policies (UniPi) and learn from sim, real, and web-scale videos…

2023-02-05 01:12:13 How are we doing in computing wrt this summary in June 2022? On algorithms, any great new ideas like FlashAttention was back then? https://t.co/OBFzpmR9IY

2023-02-05 00:55:03 RT @arankomatsuzaki: Multimodal Chain-of-Thought Reasoning in Language Models Multimodal-CoT outperforms GPT-3.5 by 16% (75.17% ->

2023-02-05 00:53:50 RT @karpathy: The most dramatic optimization to nanoGPT so far (~25% speedup) is to simply increase vocab size from 50257 to 50304 (nearest…

2023-02-05 00:50:02 RT @ctnzr: Wisdom from Jensen Huang: Strategy is not about what you will do. It's about what you will sacrifice - what you will give up in…

2023-02-04 13:05:49 Nvidia released the megatron language model before the pandemic. It’s amazing how influential this paper became. A must read for people wanting to learn about AI. https://t.co/OYBayawHZa

2023-01-31 07:09:08 RT @arankomatsuzaki: Looped Transformers as Programmable Computers Presents a framework for using transformer networks as universal comput…

2023-01-31 07:06:13 RT @mathemagic1an: What if you could fit an *entire codebase* in an LLM? "Efficiently Scaling Transformer Inference" (11/2022) https://…

2023-01-31 07:05:17 RT @arankomatsuzaki: REPLUG: Retrieval-Augmented Black-Box Language Models REPLUG with the tuned retriever significantly improves the perf…

2023-01-30 01:00:00 CAFIAC FIX

2023-01-25 23:59:10 RT @Khipu_AI: We are excited to announce our silver sponsors for #Khipu2023! Thank you to Tryolabs, ASAPP, Meta and NAACL for your suppo…

2023-01-25 23:58:52 RT @arankomatsuzaki: A Watermark for Large Language Models Proposes a watermarking framework for proprietary language models and a statist…

2023-01-25 09:41:52 This video on distributed neural net training should be part of every machine learning &

2023-01-24 23:44:01 How gut bacteria are controlling your brain —fascinating https://t.co/wzh3ebTJeo

2023-01-24 22:51:06 RT @AIMSacza: A dream come true! With the support of @DeepMind, we are launching a new AI for Science Master's program at AIMS South Africa…

2023-01-23 22:43:45 RT @FeryalMP: I’m super excited to share our work on AdA: An Adaptive Agent capable of hypothesis-driven exploration which solves challengi…

2023-01-23 22:04:27 RT @emilio__ferrara: I am 5 years late but I just read this awesome ⁦@CACMmag⁩ paper by ⁦@hannawallach⁩ and it couldn’t be a more timely an…

2023-01-23 22:00:07 RT @rlfromlux: Elated that my paper “DEP-RL: Embodied Exploration for Reinforcement Learning in Overactuated and Musculoskeletal Systems” w…

2023-01-23 21:49:20 RT @JayAlammar: Despite the Generative AI craze, one of the most exciting and reliably useful areas of AI is not generative at all. It is…

2023-01-23 06:58:54 RT @he_yi_hui: I made a chatbot that controls browser by writing Selenium code at @scale_AI #GenerativeAI hackathon with @Tiancaixinxin. Th…

2023-01-22 11:05:25 RT @DinuMariusC: We are excited to present our work, combining the power of a symbolic approach and Large Language Models (LLMs). Our Symbo…

2023-01-22 10:54:13 RT @A__Diack: Akagera National Park, Rwanda, 2009. Add that to your #ICLR2023 to do list. https://t.co/pA9RjrjvRy

2023-01-22 10:50:05 RT @pfau: Our paper on Wigner crystallization with FermiNet was just published at Physical Review Letters, and selected as an Editor's Sugg…

2023-01-19 09:15:47 RT @mvladymyrov: ML algorithms need lots of data and are prone to catastrophic forgetting. We present a new method for continual few-shot l…

2023-01-19 09:15:13 RT @mathemagic1an: Clever (and easy!) trick for better LLM context retrieval for those who haven't seen it: HyDE: Hypothetical Document Em…

2023-01-19 09:09:39 @buitengebieden @jasmineweidenb1 Happy birthday Sander @buitengebieden - Your posts give us joy - the whole family loves them!

2023-01-19 09:05:56 RT @danfei_xu: Somehow I'm more impressed by the throw motion than the 540 flip. The whole thing is so freakin' awesome!

2023-01-19 09:05:25 RT @neuro_kim: A Comp Neuro &

2023-01-18 08:41:25 RT @tri_dao: I’ve been working with @AdeptAILabs and we’ve made FlashAttention even faster for long sequences! For seqlen 8K, FlashAttentio…

2023-01-18 08:36:36 RT @karpathy: New (1h56m) video lecture: "Let's build GPT: from scratch, in code, spelled out." https://t.co/2pKsvgi3dE We build and tra…

2023-01-17 06:58:28 @yudapearl @eliasbareinboim Thank you.

2023-01-17 06:57:41 RT @yudapearl: @NandoDF @eliasbareinboim Given that my neurons are enslaved to other neurons, I, too, lack agency. Still, I have the illusi…

2023-01-17 06:55:06 RT @arankomatsuzaki: Decoding of PaLM is only ~2x as costly as that of encoding seqs (in terms of per-token TPU-hours). Kinda neat autore…

2023-01-16 22:10:37 @OriolVinyalsML @depthsofwiki Choclo

2023-01-16 22:03:02 RT @martypute: Revised drafts of Chapters 2-4 of our forthcoming MDP/RL book are now available at: https://t.co/U25jPmgt7L Any feedback wou…

2023-01-16 21:57:30 @BayesianUpdater @yudapearl @eliasbareinboim @anilkseth Thanks, this is insightful. If you have papers expanding on this I’d love to read them.

2023-01-16 21:44:29 RT @SusanDavid_PhD: If you need to put in a long day at the office to complete a project, something as simple as sending a quick text or em…

2023-01-16 21:39:15 @BayesianUpdater @yudapearl @eliasbareinboim @anilkseth Brilliant, would love to hear your thoughts on this!

2023-01-16 21:37:37 @_timharley @egrefen @hausman_k https://t.co/uzADMxzMhL

2023-01-16 21:21:38 @BayesianUpdater @yudapearl @eliasbareinboim @anilkseth Haha small world. Nice to meet you.

2023-01-16 21:19:07 @BayesianUpdater @yudapearl @eliasbareinboim Online RL agents make interventions with a goal. I believe this is enough, but I’d like to hear @eliasbareinboim expanding on this as I believe he’s put a lot of thought into it. I also think multimodality helps, based on the rubber hand illusion, e.g. @anilkseth

2023-01-16 21:08:26 @yudapearl Existing chatbots are not causal reasoners. They lack agency. They treat their actions (what they said) as observations and not as interventions. Question: how do we fix this? @yudapearl I think there’s a few solutions, but I’d love to hear you thoughts. @eliasbareinboim

2023-01-16 20:59:15 RT @_akhaliq: Region-Aware Diffusion for Zero-shot Text-driven Image Editing github: https://t.co/gXTihVYgL7 https://t.co/DhAnRcJgbE

2023-01-16 20:56:45 RT @JayAlammar: AI Art Explained: How AI Generates Images https://t.co/vLYQqqE36k New video! If you want to know how AI generation work…

2023-01-14 12:08:21 RT @G_P_Needham: We're rocking CO2 for the planet.

2023-01-12 18:08:16 The bitter lesson is starting to appear in robotics, via @hausman_k https://t.co/8jLypO4IUX

2023-01-12 11:04:41 RT @taylorhowell: my @DeepMind project is out! MuJoCo MPC (MJPC) is an interactive tool for real-time behavior synthesis with predictive co…

2023-01-12 11:03:16 RT @Khipu_AI: We're thrilled that @Google is a platinum sponsor for Khipu! Google's mission is to organise the world's information &

2023-01-12 11:01:31 RT @OriolVinyalsML: Many people mistakenly think that large language models that generate one word at a time is the end game. Connecting LL…

2023-01-12 10:53:43 First day here after a family break in Argentina. Excited about @kbeguir’s Tweet. I’m so proud of the amazing positive work of @instadeepai on building AI tools to cure disease. They’ve also shown the world how an African startup can be simply the best @DeepIndaba https://t.co/mBkaanm1YW

2022-12-23 18:38:58 @Sparkyparky82 Happy holidays Kate!

2022-12-22 21:03:26 RT @DeepIndaba: Thanks for joining us on this trip down memory lane. Whether we have met, or we will meet soon, there is so much more we ca…

2022-12-22 14:11:08 A great honour to see AlphaCode and AlphaTensor mentioned among these exceptional science achievements for 2022! Very proud of my colleagues at @DeepMind https://t.co/jy8J2Zd9fr

2022-12-20 11:41:25 RT @arankomatsuzaki: The case for 4-bit precision: k-bit Inference Scaling LawsShows that 4-bit precision is almost universally optimal f…

2022-12-20 11:39:22 RT @_akhaliq: Scalable Diffusion Models with Transformersabs: https://t.co/RlOulZLZ1U largest DiT-XL/2 models outperform all prior diffu…

2022-12-20 11:39:10 RT @arankomatsuzaki: Natural Language to Code Generation in Interactive Data Science Notebooks- Builds ARCADE, a benchmark of 1,082 code…

2022-12-20 11:37:36 RT @arankomatsuzaki: Evaluating Human-Language Model InteractionFinds that non-interactive performance does not always result in better h…

2022-12-18 21:45:28 Salta, Argentina Happy people celebrating after one of the most entertaining and dramatic finals! https://t.co/0HThSuYPjN

2022-12-18 19:07:16 RT @sundarpichai: One of the greatest games ever. Well played Argentina and France. Jogo Bonito. Nobody deserves it more than #messi, imho…

2022-12-18 17:57:33 A la plaza a celebrar

2022-12-18 14:36:02 Comienza la navidad en Argentina Let’s hope Messi brings the party home! https://t.co/zfiws2tbFp

2022-12-18 09:38:23 @shaneguML This is hilarious

2022-12-15 10:43:10 RT @yuvaltassa: Yours truly, delivering the MIT Robotics Seminar about our new MJPC tool.https://t.co/pu7mmyMkcN

2022-12-13 21:55:52 I feel so fortunate to live through some of the most transformative technological advances in history. AI, biotechnology, electric cars, landing rockets, the inception of fusion … amazing! Let’s put it to the benefit of human kind! https://t.co/NKLRiP9rQl

2022-12-13 21:38:48 Vamos!! https://t.co/Vw4lGjFxv1

2022-12-12 20:14:42 RT @_akhaliq: VindLU: A Recipe for Effective Video-and-Language Pretrainingabs: https://t.co/paA78x3Ucxcode: https://t.co/2OL4bmXPU1 http…

2022-12-12 09:18:19 RT @DavidDeutschOxf: Trying the Socratic method on ChatGPT: https://t.co/fksDaJlS9P

2022-12-12 09:16:28 RT @DaveJuergens: We’re very happy to announce that our RFdiffusion manuscript is now on bioRxiv! A lot can change in a week - we’ve now te…

2022-12-12 09:14:01 RT @tfgg2: Pleased to say @DeepMind has just released a code update to #AlphaFold 2.This includes updated model weights trained on newer…

2022-12-12 09:13:05 @kayo_yin @emnlpmeeting @gneubig Congratulations

2022-12-12 09:12:57 RT @kayo_yin: We won #EMNLP2022 Honorable Mention Long Paper!!! Thank you so much @emnlpmeeting for the recognition, I’m so excited and h…

2022-12-12 09:11:17 RT @arankomatsuzaki: We have released "Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints"!Our method converts a pretra…

2022-12-12 09:10:16 RT @JayAlammar: Adding Human Intelligence to ML models with human-learn by @fishnets88 Clip: https://t.co/y217RKqjuiLibrary: https://t.c…

2022-12-12 09:09:47 @JayAlammar @yoavgo @kayo_yin @gneubig Congratulations

2022-12-12 09:09:27 RT @yoavgo: congrats to @kayo_yin and @gneubig for winning the EMNLP 2022 Long Paper Honorable Mention award. https://t.co/JPnxKIOtPS

2022-12-10 18:44:13 RT @DeepMind: Congratulations to this year’s #CASP15 winners and participants for their achievements! In the two years since #CASP14, we…

2022-12-09 09:28:52 RT @arankomatsuzaki: VideoDex: Learning Dexterity from Internet VideosOutperforms various SotA methods on various manipulation tasks.pr…

2022-12-09 09:25:01 RT @CovariantAI: This week at @FortuneMagazine's #BrainstormAI, Covariant’s own, @pabbeel explains our approach to building our generalized…

2022-12-09 09:21:36 RT @liyuajia: Glad to share that our work #AlphaCode from earlier this year is now published in Science as the front cover! Proud of what…

2022-12-08 22:42:35 RT @CollinBurns4: How can we figure out if what a language model says is true, even when human evaluators can’t easily tell?We show (http…

2022-12-08 22:41:08 RT @_akhaliq: Diffusion-SDF: Text-to-Shape via Voxelized Diffusionabs: https://t.co/KZdqcqKGZP https://t.co/gSfHVLTeq5

2022-12-08 22:31:35 Proud to see AlphaCode published on @ScienceMagazine and above all proud of the AlphaCode team for their inspiring and pioneering work, and for their collaborative spirit and inclusive practices. https://t.co/lWHmHoR9Yd

2022-12-08 22:15:53 Impressive to see a generalist neural net implementation of so many of the classical algorithms taught in undergraduate computer science. https://t.co/AT1FqcMGpZ

2022-12-08 22:02:51 RT @ScienceMagazine: #AlphaCode—a new #AI system for developing computer code developed by @DeepMind—can achieve average human-level perfor…

2022-12-07 16:15:08 RT @arankomatsuzaki: InternVideo: General Video Foundation Models via Generative and Discriminative LearningPresents general video founda…

2022-12-07 16:14:27 RT @NatureComms: .@yorambac @JanosKramar @drimgemp @a_tacchetti @empiricallykev @MateuszOnAI @ThoreG et al. report how to make intelligent…

2022-12-07 16:11:56 RT @DeepMind: Mutations in KCTD proteins are linked to a range of diseases, including schizophrenia and leukaemia, and yet many of these pr…

2022-12-07 00:38:40 RT @_akhaliq: PhysDiff: Physics-Guided Human Motion Diffusion Modelabs: https://t.co/lST8NvILxu project page: https://t.co/iwEEK1TOzX htt…

2022-12-07 00:37:43 RT @arankomatsuzaki: Meta-Learning Fast Weight Language ModelsPresents Fast Weight Layers (FWLs), a neural component that provides the be…

2022-12-07 00:06:53 RT @jacobandreas: Speculative (!!!) paper arguing that big LMs can model agency &

2022-12-05 23:47:55 @wooldridgemike @CompSciOxford @demishassabis @DeepMind You’re welcome

2022-12-05 23:47:29 RT @wooldridgemike: These are the toughest times in the tech sector for years - it’s wonderful to that DeepMind are able to continue with t…

2022-12-05 08:55:47 @kchonyc You’re bigger than life!

2022-12-04 13:12:06 New robotics task? https://t.co/exVFjAPIlT

2022-12-03 12:35:45 RT @ScienceMagazine: A newly developed #AI agent called DeepNash learned to play Stratego—one of the few board games AI has not yet mastere…

2022-12-02 11:48:32 @aidangomezzz You’ll need surgery again after that Enjoy, and heal

2022-12-01 14:46:43 RT @bentossell: All the best examples of ChatGPT, from OpenAI:

2022-12-01 13:45:31 RT @DeepMind: Work by: @summerfieldlab, @hannahsheahan, @mhtessler, @MartinJChadwick and @bakkermichiel.

2022-12-01 01:01:52 RT @summarizedml: A growing ecosystem of large, open-source foundation models has reduced the costs of building both harmful and beneficial…

2022-12-01 00:30:41 RT @amasad: ChatGPT could be a good debugging companion

2022-12-01 00:29:23 This application of generative models is one of the most impressive AI demos I’ve ever seen. I’m amazed at the results but also at how unexpectedly easy, creative and accessible these tools have become. We do live in special times. https://t.co/a01VlbqBNv

2022-11-30 23:23:28 RT @GuyP: OK so @OpenAI's new #ChatGPT can basically just generate #AIart prompts. I asked a one-line question, and typed the answers verba…

2022-11-30 23:14:14 RT @OpenAI: Try talking with ChatGPT, our new AI system which is optimized for dialogue. Your feedback will help us improve it. https://t.c…

2022-11-30 09:33:59 Transforming neural nets as hardware evolves is transformative. @ilyasut https://t.co/WirmYbT3zk

2022-11-30 09:22:24 @aidangomezzz Wishing you a healthy recovery

2022-11-30 09:19:18 RT @laurentsifre: Join us at the Chinchilla poster tomorrow to discuss LLMs and compute optimal scaling!Wed 30 Nov 4:30 p.m. CST — 6 p.m.…

2022-11-30 09:19:00 @kchonyc Time to give JAX a try! Irresistible

2022-11-29 21:23:06 RT @alexgkendall: I’m often asked how does @wayve_ai use simulation to develop our next-generation autonomous driving?Today we’re unveili…

2022-11-29 21:19:52 RT @du_yilun: Introducing Decision Diffuser, a conditional diffusion model that outperforms offline RL across standard benchmarks – using o…

2022-11-29 21:19:30 RT @GoogleAI: As the scale of models increases, the computational effort needed for training grows rapidly. Check out a pair of new approac…

2022-11-29 21:18:56 RT @DeepMind: Visit the Playhouse at #NeurIPS2022! In this 3D virtual world, you'll be able to communicate with a multimodal interactive ag…

2022-11-29 21:17:04 @feishaAI @ZoubinGhahrama1 Was thinking the same, 25

2022-11-29 07:09:00 RT @sedielem: New paper: continuous diffusion for categorical dataWe train diffusion language models with cross-entropy, using score inte…

2022-11-28 23:02:05 RT @StableDiffusion: Happy to share this HUGE announcement! SD 2.0 for all! Comes with text-to-image, depth-to-image, inpainting, and u…

2022-11-28 22:58:16 RT @rsalakhu: Causal Confounds in Sequential Decision Making – Machine Learning Blog | ML@CMU | Carnegie Mellon University https://t.co/6KS…

2022-11-28 22:49:30 RT @yanndubs: #NeurIPS2022What are ideal representations for self-sup. learning (SSL)?We give simple optimality conditions and use them…

2022-11-28 22:46:16 RT @mengjiao_yang: See you all at the 1st Foundation Models for Decision Making workshop @NeurIPSConf (Room 391) on Sat, Dec 3 2022. See sc…

2022-11-28 22:43:44 RT @GoogleAI: Stop by our booth today at 4pm to have a video chat with a robot arm! Ask it to move around objects, or just say hello. Power…

2022-11-28 22:43:06 RT @DeepMind: How can we get language models to solve maths problems accurately with correct, human-interpretable reasoning?We evaluate m…

2022-11-28 07:09:41 RT @The_Numbat: Two SIGGRAPH best papers covered neural graphics primitives: networks that encode signals like images, surfaces, and light…

2022-11-28 00:07:01 Guess the prompt behind these AI-generated images - fun game! ⁦@le_roux_nicolas⁩ https://t.co/bACTRVyCkU

2022-11-27 23:49:09 “Approximate Bayesian computation with deep learning supports a third archaic introgression in Asia and Oceania” - fascinating application. https://t.co/lfjYQ8SlT6

2022-11-27 18:04:17 RT @DeepMind: Looking for a workshop at #NeurIPS2022? Join us tomorrow: @WiMLworkshop - 9am-12pm CST @black_in_ai - 9-11am CST @_…

2022-11-26 23:17:53 RT @lexfridman: Here's my conversation with Guido van Rossum (@gvanrossum), creator of Python, one the most popular and impactful programmi…

2022-11-26 23:14:07 RT @anammostarac: “Everything around you that you call life was made up by people that were no smarter than you and you can change it, you…

2022-11-26 23:10:55 RT @MIT_CSAIL: 9 key computer science topics - and the best book and video for learning each of them: https://t.co/i6qNiE8qhr (credit @Brad…

2022-11-26 10:21:16 When will our large neural network models invent one of these? What will it be?! I dream of this novel knowledge creation, because the generation of symbolic abstractions and descriptions of the world has been great for generalisation beyond the senses. https://t.co/w8wpZnm7KT

2022-11-25 12:10:02 RT @DeepMind: Announcing the next version of Melting Pot, our evaluation platform for multi-agent AI research.Melting Pot 2.0 tests wheth…

2022-11-25 12:09:42 RT @JayAlammar: Big update to "The Illustrated Stable Diffusion" posthttps://t.co/sbjKHP5ndT14 new and updated visuals. The biggest u…

2022-11-25 08:14:06 RT @XingyouSong: I'm excited to present the OptFormer at @NeurIPSConf 2022 (Wednesday 11:30 - 1 PM at Hall J, #129)!Would love to talk ab…

2022-11-24 10:12:52 RT @CohereAI: Build Chatbots Faster with Large Language Modelshttps://t.co/2b2NQTspYYIn this video, Dr. Rachael Tatman (@rctatman) walks…

2022-11-24 10:10:21 RT @arankomatsuzaki: Retrieval-Augmented Multimodal Language ModelingPresents RA-CM3, which exhibits novel capabilities like knowledge-in…

2022-11-22 23:57:02 @SuryaGanguli @sshkhr16 @arimorcos Congratulations

2022-11-22 23:56:56 RT @SuryaGanguli: Our "Beyond Neural Scaling laws" paper got a #NeurIPS22 outstanding paper award! Congrats Ben Sorscher, Robert Geirhos, @…

2022-11-22 07:05:59 @yukez Congratulations!

2022-11-21 22:33:56 RT @Yuhu_ai_: Language models can dramatically improve their reasoning by learning from chains of thought that they generate. With STaR,…

2022-11-21 22:31:04 RT @_akhaliq: RenderDiffusion: Image Diffusion for 3D Reconstruction, Inpainting and Generationabs: https://t.co/FnqzvKa6sQ https://t.co/W…

2022-11-21 22:30:45 RT @urialon1: New Paper: Program-aided Language modelsPrompting methods such as chain-of-thought (@_jasonwei) employ LLM for decomposin…

2022-11-21 22:22:08 RT @GoogleAI: Applications for the first-ever Google PhD Fellowships for students in Latin America open today, along with applications to s…

2022-11-18 12:05:25 RT @jaschasd: If there is one thing the deep learning revolution has taught us, it's that neural nets will outperform hand-designed heurist…

2022-11-18 12:04:59 RT @percyliang: Language models are becoming the foundation of language technologies, but when do they work or don’t work? In a new CRFM pa…

2022-11-17 10:51:28 RT @GoogleAI: Introducing a novel mixture-of-experts routing algorithm, called Expert Choice, that can achieve optimal load balancing betwe…

2022-11-17 10:50:15 RT @ramin_m_h: In a new article published today in Nature MI @NatMachIntell we solved a differential equation that describes the interactio…

2022-11-17 10:49:10 RT @karpathy: Good post. A lot of interest atm in wiring up LLMs to a wider compute infrastructure via text I/O (e.g. calculator, python in…

2022-11-17 10:48:13 RT @AIMS_Next: We’re proud to help build a stronger, diverse and more inclusive AI community for the future through the @DeepMind global sc…

2022-11-15 22:22:35 RT @docmilanfar: Why don’t more people know about the gem that is Tweedie's formula?Say is a noisy measurement of = + w/ ∼…

2022-11-15 22:21:01 RT @SusanDavid_PhD: Before you act, breathe into the space between stimulus and response, and make the choice to be the person you most wan…

2022-11-15 22:11:06 RT @wuyusongwys: We are glad to announce our new work collaborating with https://t.co/bhnCPtfly8: a large-scale text-audio contrastive lear…

2022-11-15 22:10:49 RT @JitendraMalikCV: Our robot dog can go up and down stairs, walk on stepping stones where even a single bad foot placement would lead to…

2022-11-15 22:09:40 RT @ylecun: A Large Language Model trained on scientific papers.Type a text and https://t.co/XKTkxs8Ae0 will generate a paper with relevan…

2022-05-20 01:00:00Cafiac Fix

2001-01-01 01:01:01

Découvrez Les IA Experts

Nando de Freitas Researcher at Deepind
Nige Willson Speaker
Ria Pratyusha Kalluri Researcher, MIT
Ifeoma Ozoma Director, Earthseed
Will Knight Journalist, Wired