Découvrez Les IA Experts
 |
Nando de Freitas | Chercheur chez Deepind |
 |
Nige Willson | Conférencier |
 |
Ria Pratyusha Kalluri | Chercheur, MIT |
 |
Ifeoma Ozoma | Directrice, Earthseed |
 |
Will Knight | Journaliste, Wired |
Profil AI Expert
Non Disponible
Les derniers messages de l'Expert:
2024-11-15 23:13:13 RT @CoLLAs_Conf: Exciting News! The Fourth Conference on Lifelong Learning Agents (CoLLAs 2025) will be held at the University of Pennsyl…
2024-11-10 17:29:25 My profile on Bluesky https://t.co/LssDD0zxsB
2024-11-08 20:10:29 @benczheng Interesting, any links?
2024-11-08 16:52:06 @RediZypce Paper showing it led to increased prices as tariffs were passed to consumers https://t.co/OixTjhJ9f4
2024-11-08 16:50:18 @RediZypce He had to bail out farmers https://t.co/zb5N6OMdkq
2024-11-08 16:34:23 @RediZypce Trump has not specified really anything, but he's at times claimed he wants broad tariffs in whole industries, I don't see why that wouldn't raise prices of groceries. Everything is global now, to undo that would take decades.
2024-11-08 16:20:57 @RediZypce Also I kind of don't want the government to skimp on food regulations, not sure there's a good profit to be made in that kind of thing, so why would the private sector take that up? Same for health and safety
2024-11-04 23:22:28 No matter what the outcome, the Harris campaign has been superb, and a testament to what can be accomplished by passionate, inspired people fighting for freedom, hope, and our futures. If you can, please vote tomorrow!
2024-10-13 12:06:31 @howdataworks Mostly just the podcast feature, and also the template notes it creates, like study guides and FAQs. Really remarkable how it can pull out specific details from a lot of sources
2024-10-12 16:34:14 Kind of shocked at how good https://t.co/pnkblUvZNX is, especially the podcast generation feature. Getting so much insight into my past writings, and it's also helpful for literature review. Honestly it's one of the best use cases of LLMs I've seen!
2024-05-15 00:51:25 RT @mmmbchang: Gemini and I also got a chance to watch the @OpenAI live announcement of gpt4o, using Project Astra! Congrats to the OpenAI…
2024-05-05 16:39:16 RT @juliancodaforno: 1/ Excited to share our new paper "CogBench: a large language model walks into a psychology lab" accepted at #ICML20…
2024-04-18 16:40:59 RT @schwarzjn_: Announcing the next Harvard #efficientml seminar on Learning to learn, featuring @janexwang and Rising Star @juliancodafo…
2024-04-10 12:54:59 RT @NASAMoon: Oops I did it again #TotalSolarEclipse https://t.co/JXPe26qq3Q
2024-03-22 20:47:52 Longer context is a game changer, now we're going to see the real power of in-context (meta-) learning! https://t.co/YbSItaIwUh
2024-03-01 00:00:00 CAFIAC FIX
2024-03-11 00:00:00 CAFIAC FIX
2023-05-19 19:00:00 CAFIAC FIX
2023-05-21 19:00:00 CAFIAC FIX
2023-04-21 00:00:01 CAFIAC FIX
2023-04-18 19:24:25 RT @cpilab: Happy that our paper on meta-learned theories of cognition has finally been submitted. Joint work with Marcel Binz, Ishita Dasg…
2023-03-28 00:03:54 @ChrisMurphyCT Please listen to all of the expert ML researchers commenting on this and saying it's just plain wrong. We've got a lot of work to do to better understand LLMs, but this kind of rhetoric is simply not helpful.
2023-03-27 23:45:34 ChatGPT is wildly confident, but so wrong https://t.co/D0gbIM4CZt
2023-03-27 23:41:23 The comments section on this post has some interesting insights, but it'd be nice to see a systematic analysis. How accurate are these confidence scores? https://t.co/o3cRHzvUtH
2023-03-27 23:35:54 Interesting - asking GPT-4 about its confidence allows it to be more accurate about generating real references, while ChatGPT in my experience rarely generated real papers. Better calibration? (also have I missed my calling with long-tailed RL?) https://t.co/HrF3bkIUOU https://t.co/DrnP8W472z
2023-03-05 11:59:52 @gershbrain You were definitely there in spirit, although missed. I even referenced your talk in my talk!
2023-03-05 10:00:00 CAFIAC FIX
2023-03-02 22:00:00 CAFIAC FIX
2023-02-27 01:00:00 CAFIAC FIX
2023-02-17 21:11:42 I like you AI/ML twitter, but this week has made me absolutely exhausted. Here's something about black holes! https://t.co/3DBAVf8HG2
2023-02-16 16:13:20 @amirbolous You can search for past reddit responses, which is then instantaneous. It's possible that the response of chatGPT is essentially doing an implicit search of reddit answers it's trained on, and I'd rather just use google for that
2023-02-12 01:42:40 @raphaelmilliere As far as I know, no one has explored this - it's an open empirical research question! So I'm not saying I completely agree with him, but this insightful point in Ted's essay seems to have been missed in favour of his lossy compression analogy
2023-02-12 01:37:35 @raphaelmilliere If the generated text is truly original, then it should improve the model just as much as human created data, and there'd be no problem training on internet data which might include massive amounts of its own output, as these LLMs get more popular
2023-02-12 01:34:15 @raphaelmilliere The question of what is truly original is a tough one and is clearly inspiring a lot of debate! IMO what's critical is whether the model output would itself make good training data for the next iteration of the model.
2023-02-11 01:14:06 Of course we don't know yet that you can't train a model on its own outputs, see eg https://t.co/1OXIWRXoYn, but this is still an open question. This paper is also not directly training on the model's raw outputs, but uses some additional metric of output quality to do filtering
2023-02-11 01:03:49 I think this is a misreading of what Ted Chiang is saying. He's not claiming that chatGPT does literal blurring but just in text space. He's saying that there's no original content creation, only remixing, hence why you wouldn't want to train a LLM on its own (unfiltered) outputs https://t.co/XY1CWMNQR5
2023-02-10 20:48:08 RT @TimKietzmann: Just finished my course “Machine Learning for Cognitive Computational Neuroscience”. Across 12 lectures (90 minutes each)…
2023-02-10 20:47:47 RT @jathansadowski: As much as Ted Chiang's fiction is fantastic, it's really his non-fiction that I could read forever. His two paragraphs…
2023-01-30 01:00:00 CAFIAC FIX
2023-01-21 01:39:00 RT @FeryalMP: I’m super excited to share our work on AdA: An Adaptive Agent capable of hypothesis-driven exploration which solves challengi…
2022-12-16 00:45:20 @robinhanson Audio-visual equipment that doesn't break right before your big talk.
2022-12-16 00:41:14 I don't check Twitter for two days and I swear there are about 10 major announcements about some crazy new thing a large-scale model can do. It's incredibly exciting but I wonder what we're sacrificing in all this rush
2022-12-09 20:45:45 ChatGPT wasn't as successful explaining how RLHF works. Perhaps they will use this blog post to further fine tune it via RLHF, thus leading it to better explain itself from considering its own output, and I find that haunting. https://t.co/Ivmhd4gMpe
2022-12-09 20:35:31 Massive datasets are important for LLMs like #ChatGPT, but what gets lost is that human feedback is crucial! Great blog post explaining RLHF https://t.co/uaBfBIi9gD An even more sophisticated version of fine tuning LLMs from human feedback is work done by https://t.co/BYbkigZXn5
2022-12-09 18:14:26 @drjwrae take!
2022-12-08 13:00:00 CAFIAC FIX
2022-12-07 08:00:00 CAFIAC FIX
2022-11-11 13:19:02 The biggest failure of social media is that they allow single individuals to completely wreck our public discourse. We need built in failsafes and regulation - it should be a public utility, not a revenue stream
2022-10-24 22:26:37 This is why generalisation and fast adaptation to new tasks are the only true benchmarks. Life never puts you in the exact same situation twice, so why do we keep classifying the same 10 digits / 1000 classes of dogs? https://t.co/CNSicE6NB6
2022-10-17 15:08:09 RT @summerfieldlab: I think it's probably time to share this...I wrote a book. It's coming out in dec 2022 (in theory). It is intended to h…
2022-10-13 18:07:07 RT @NeurIPSConf: NeurIPS is creating a new high-school outreach day in New Orleans on Monday, Nov 28. Our vision is to empower students to…
2022-10-13 18:03:01 @Sagar_Vaze @TaoYu10 Sorry I wasn't trying to call you out or anything - it's not great as a reviewer to feel like your hard work isn't being taken into account, but at the same time the current system is susceptible to gatekeeping. Hard to find a good balance, and everyone is overloaded...
2022-10-13 17:56:14 RT @scychan_brains: New paper : Transformer inductive biases!Transformers generalize differently from information stored in:‣ weights -…
2022-10-13 12:58:09 @Sagar_Vaze This is a bit disturbing, as it implies you kept bidding for it at consecutive conferences. Were no changes made at all between submissions, or were just your changes not incorporated?
2022-10-07 17:26:01 Meta-reinforcement learning what you don't know - and how much you don't know it - super cool work by my colleagues! https://t.co/CeX7gTSWMt
2022-10-06 15:40:45 RT @scychan_brains: We've released the codebase for the paper "Data Distributional Properties Drive Emergent In-Context Learning in Transfo…
2022-10-05 11:11:38 The total number of AI papers might be growing exponentially, but not the total number of AI ideas - there is surely a lot of redundancy, especially since human attention and capacity for information processing is fixed. Hence all these public arguments about credit assignment. https://t.co/x0NVpiIOaK
2022-09-16 13:58:02 Applications open TODAY for DeepMind 2023 internships, come join and work with us!https://t.co/uVBbcTqEW0 (due Sept 30) https://t.co/DT2NqdhGoS
2022-09-16 13:55:05 @ericjang11 Surely it's to be expected that among all batches of N papers (N being <
2022-07-09 12:17:25 Looking forward to participating in the great Brain Debate today at #FENS2022 (albeit virtually due to the recent fun covid wave)! https://t.co/nlj2Wxg6dw
2022-07-07 17:50:46 RT @TheCrick: We’re excited to welcome @DeepMind to the Crick where they’ll set up a new lab to apply #AI and #machinelearning to address b…
2022-06-13 23:36:22 @tyrell_turing 0.68 on a Monday
2022-06-13 23:35:17 Why has my whole feed turned into my freshman year philosophy class?
2022-06-10 09:00:11 RT @akjagadish: I will be presenting our work “Probing Compositional Inference in Natural and Artificial Agents” as a poster (P.67) today i…
2022-06-07 10:37:29 @ermgrant @SWC_Neuro @GatsbyUCL @SaxeLab Welcome to London - we should grab coffee sometime!
2022-06-02 07:21:49 @HenrikMettler These models are a bit black box currently. AI analytics was already a thing, but as models get more sophisticated it's just looking more and more like the kind of testing we do with humans and animals
2022-06-01 12:14:11 ... which would indicate that such structure is inherent in the data itself
2022-06-01 12:13:37 This is going to be an entire new subfield - the psychology and cognitive science of large-scale models. These particular findings are probably incidental, but it'd be fascinating to discover what findings hold across models, and architectures... https://t.co/5aT8rvVGfr
2022-05-27 21:59:57 RT @LindauerMarius: To accommodate journal track submissions, we extended the early-bird rates of our #AutoMLConf a bit until May 27th. htt…
2022-05-20 08:11:00 CAFIAC FIX
2022-10-24 22:26:37 This is why generalisation and fast adaptation to new tasks are the only true benchmarks. Life never puts you in the exact same situation twice, so why do we keep classifying the same 10 digits / 1000 classes of dogs? https://t.co/CNSicE6NB6
2022-10-24 22:26:37 This is why generalisation and fast adaptation to new tasks are the only true benchmarks. Life never puts you in the exact same situation twice, so why do we keep classifying the same 10 digits / 1000 classes of dogs? https://t.co/CNSicE6NB6
2022-10-24 22:26:37 This is why generalisation and fast adaptation to new tasks are the only true benchmarks. Life never puts you in the exact same situation twice, so why do we keep classifying the same 10 digits / 1000 classes of dogs? https://t.co/CNSicE6NB6
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-25 17:40:27 The upshot is that we will always need artists and humans to create truly new art and inject diversity, because AI will always tend toward mode collapse
2022-11-25 17:38:46 Since these huge models are trained on internet data, and more and more of their generated output is being used and put on the internet, they will be increasingly trained on their own output, which will lead to them generating art which appears to be more and more derivative.
2022-11-25 17:38:45 It’s true that generative AI will become a powerful tool for being able to create graphic representations, for both artists and non-artists, but it won’t replace human creativity. At their core, these models can interpolate and recombine, but they can’t create new meaning.
2022-11-25 17:38:44 Being on #Thanksgiving vacation = having time to randomly think about AI! Regarding the argument that AI will replace artists: I have some thoughts about this.
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-26 22:52:25 @awjuliani Good point! Also possibly the price per ad has decreased due to lower demand, which leads to more ads?
2022-11-26 15:29:02 What strategy is post-Elon takeover Twitter going for? I’m detecting an increase in off-topic, but viral tweets (in categories like “funny”, “cats”, anything to do with Elon), and noticeably more ads.
2022-11-26 13:57:00 Heading to #NeurIPS2022 in New Orleans! I'll be at the DeepMind booth on Monday at 5 pm and Thursday at 10 am, and mentoring at NeurIPS Education Outreach Day on Monday (11 am). If you're around please say hi!
2022-11-25 17:40:27 The upshot is that we will always need artists and humans to create truly new art and inject diversity, because AI will always tend toward mode collapse
2022-11-25 17:38:46 Since these huge models are trained on internet data, and more and more of their generated output is being used and put on the internet, they will be increasingly trained on their own output, which will lead to them generating art which appears to be more and more derivative.
2022-11-25 17:38:45 It’s true that generative AI will become a powerful tool for being able to create graphic representations, for both artists and non-artists, but it won’t replace human creativity. At their core, these models can interpolate and recombine, but they can’t create new meaning.
2022-11-25 17:38:44 Being on #Thanksgiving vacation = having time to randomly think about AI! Regarding the argument that AI will replace artists: I have some thoughts about this.
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-27 19:40:52 @BeingMIAkashs See you there!
2022-11-26 22:52:25 @awjuliani Good point! Also possibly the price per ad has decreased due to lower demand, which leads to more ads?
2022-11-26 15:29:02 What strategy is post-Elon takeover Twitter going for? I’m detecting an increase in off-topic, but viral tweets (in categories like “funny”, “cats”, anything to do with Elon), and noticeably more ads.
2022-11-26 13:57:00 Heading to #NeurIPS2022 in New Orleans! I'll be at the DeepMind booth on Monday at 5 pm and Thursday at 10 am, and mentoring at NeurIPS Education Outreach Day on Monday (11 am). If you're around please say hi!
2022-11-25 17:40:27 The upshot is that we will always need artists and humans to create truly new art and inject diversity, because AI will always tend toward mode collapse
2022-11-25 17:38:46 Since these huge models are trained on internet data, and more and more of their generated output is being used and put on the internet, they will be increasingly trained on their own output, which will lead to them generating art which appears to be more and more derivative.
2022-11-25 17:38:45 It’s true that generative AI will become a powerful tool for being able to create graphic representations, for both artists and non-artists, but it won’t replace human creativity. At their core, these models can interpolate and recombine, but they can’t create new meaning.
2022-11-25 17:38:44 Being on #Thanksgiving vacation = having time to randomly think about AI! Regarding the argument that AI will replace artists: I have some thoughts about this.
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage
2022-11-27 19:40:52 @BeingMIAkashs See you there!
2022-11-26 22:52:25 @awjuliani Good point! Also possibly the price per ad has decreased due to lower demand, which leads to more ads?
2022-11-26 15:29:02 What strategy is post-Elon takeover Twitter going for? I’m detecting an increase in off-topic, but viral tweets (in categories like “funny”, “cats”, anything to do with Elon), and noticeably more ads.
2022-11-26 13:57:00 Heading to #NeurIPS2022 in New Orleans! I'll be at the DeepMind booth on Monday at 5 pm and Thursday at 10 am, and mentoring at NeurIPS Education Outreach Day on Monday (11 am). If you're around please say hi!
2022-11-25 17:40:27 The upshot is that we will always need artists and humans to create truly new art and inject diversity, because AI will always tend toward mode collapse
2022-11-25 17:38:46 Since these huge models are trained on internet data, and more and more of their generated output is being used and put on the internet, they will be increasingly trained on their own output, which will lead to them generating art which appears to be more and more derivative.
2022-11-25 17:38:45 It’s true that generative AI will become a powerful tool for being able to create graphic representations, for both artists and non-artists, but it won’t replace human creativity. At their core, these models can interpolate and recombine, but they can’t create new meaning.
2022-11-25 17:38:44 Being on #Thanksgiving vacation = having time to randomly think about AI! Regarding the argument that AI will replace artists: I have some thoughts about this.
2022-11-18 14:03:31 @ludwig_stumpp Why would I have access to traffic data? Just going off this tweet https://t.co/Uxguc9cFlJ
2022-11-18 12:40:35 I don't believe this has ever happened before - all previous social networks have died because of a declining user base, and lack of demand. It seems like twitter could still be viable, but it's just caught in a death spiral
2022-11-18 12:37:46 It's actually really surreal watching a digital universe aware and commenting on its own impending destruction, and which is ironically driving increased usage