For the year-end and New Year holidays, I've summarized a little bit of what I've been thinking about.
You can watch it for fun, empathy and objection are all welcome!
Reflection on the Conceptual Essence of Language Models, the Path to AGI
cosmoquester.github.io/reflection-on-…
# RLHF is just barely RL
Reinforcement Learning from Human Feedback (RLHF) is the third (and last) major stage of training an LLM, after pretraining and supervised finetuning (SFT). My rant on RLHF is that it is just barely RL, in a way that I think is not too widely…
1/ How do humans and animals form models of their world?
We find that Foundation Models for Embodied AI may provide a framework towards understanding our own “mental simulations”. 🧵👇
arxiv.org/abs/2305.11772
with awesome collaborators: @rishi_raj@mjaz_jazlab@GuangyuRobert
@alex_peys The problem isn't that it is a transformer.
The problem is that it is an auto-regressive LLM.
Auto-regressive LLMs that compute each token with a fixed number of computational steps can't reason, regardless of the details of the architecture.
Our paper has been accepted as a Spotlight Poster!
Congratulations @cosmoquester
Please visit our poster session and leave your comments and questions.
icml.cc/virtual/2024/p…#ICML
Our paper has been accepted as a Spotlight Poster!
Congratulations @cosmoquester
Please visit our poster session and leave your comments and questions.
icml.cc/virtual/2024/p…#ICML https://t.co/YEwaCSQ9zg
Nice article in Financial Time where I explain that Auto-Regressive LLM are insufficient to reach human-level intelligence (or even cat-level intelligence).
But alternative architectures that I call "objective driven" may reach human-level intelligence one day.
They use world…
Nice article in Financial Time where I explain that Auto-Regressive LLM are insufficient to reach human-level intelligence (or even cat-level intelligence).
But alternative architectures that I call "objective driven" may reach human-level intelligence one day.
They use world…
I am very pleased to announce my first paper "Memoria: Resolving Fateful Forgetting Problem through Human-Inspired Memory Architecture" at #ICML2024!
Memoria processes long sequence information inspired by human memory systems.
More Details: arxiv.org/abs/2310.03052
Excited to share our paper "Memoria: Resolving Fateful Forgetting Problem through Human-Inspired Memory Architecture" presented at ICML 2024!
Congratulations to Sangjun Park!
Preprint: arxiv.org/abs/2310.03052#ICML2024
I do not believe human-level AI (artificial superintelligence, or the commonest sense of #AGI) is close at hand. AI has made breakthroughs, but the claim of AGI by 2030 is as laughable as claims of AGI by 1980 are in retrospect. Look how similar the rhetoric was in @LIFE in 1970!
26 Followers 69 FollowingPh.D. student #SKKU Human Language Intelligence Lab.
Interested in natural language processing, language modeling.
Instagram:
https://t.co/Vcn035RkiD
3K Followers 120 FollowingResearch Fellow at @Harvard and incoming Asst Prof at @JohnsHopkins interested in language, cognition, and AI. Formerly: PhD @MIT.
3K Followers 2K Following🚨 I'm no longer active on Twitter.
Find me on 🦋: @[email protected]
Senior Research Fellow @GatsbyUCL & @SWC_Neuro
{learning, representations} in 🧠💭🤖
365K Followers 6K FollowingChief Scientist, Google DeepMind & Google Research. Gemini Lead. Opinions stated here are my own, not those of Google. TensorFlow, MapReduce, Bigtable, ...
5K Followers 1K FollowingLanguage and thought in brains vs machines.
New Assistant Prof @ Georgia Tech Psychology. Previously: postdoc @MIT_Quest & PhD @mitbrainandcog.
She/her
26 Followers 69 FollowingPh.D. student #SKKU Human Language Intelligence Lab.
Interested in natural language processing, language modeling.
Instagram:
https://t.co/Vcn035RkiD
12K Followers 2K FollowingAssociate Professor at Harvard & Kempner Institute. Applying computational frameworks & ML to decode multi-scale neural processes. Marathoner. Rescue dog mom.
3K Followers 1K FollowingTheoretical neuroscience, theory of neural computation, physics of learning and intelligence. Assistant Professor of Applied Mathematics @Harvard SEAS
77K Followers 2K Followinga combination of a mediocre scientist, a mediocre manager, a mediocre advisor & a mediocre physicist at @nyuniversity (@CILVRatNYU) & @PrescientDesign
4K Followers 46 FollowingDirector, MIT Quest for Intelligence and Professor of Neuroscience, @MIT. Our research goal is to reverse engineer the mechanisms of human intelligence.
6K Followers 2K FollowingCS PhD Student at Stanford Trustworthy AI Research with @sanmikoyejo. Prev interned/worked @ Meta, Google, MIT, Harvard, Uber, UCL, UC Davis
4K Followers 112 FollowingWe study human learning using tools from machine learning and improve machine learning using insights from cognitive science.
3K Followers 22 FollowingThe Center for Brains, Minds and Machines is a multi-institutional NSF Center dedicated to the study of the science and engineering of intelligence.
949K Followers 764 FollowingProfessor at NYU. Chief AI Scientist at Meta.
Researcher in AI, Machine Learning, Robotics, etc.
ACM Turing Award Laureate.
6K Followers 2K FollowingSenior research scientist at @LosAlamosNatLab. Former prof at @ucl
and @UTAustin. CogSci, AI, Comp Neuro, AI for scientific discovery
Also @profdata on Bluesky
18K Followers 4K FollowingAI professor.
Deep Learning, AI alignment, ethics, policy, & safety.
Formerly Cambridge, Mila, Oxford, DeepMind, ElementAI, UK AISI.
AI is a really big deal.
No recent Favorites. New Favorites will appear here.