Andrew Cohen
@andrew_e_cohen
Followers
211
Following
726
Media
0
Statuses
134
Reinforcement learning @AIatMeta | Previously #mlagents
Joined December 2019
RT @soniajoseph_: Our paper Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video received an Oral at the Mecā¦.
0
30
0
RT @qqyuzu: d1: to grow in reasoning, masked diffusion language models go beyond supervised learning , we meet RL !š
0
15
0
RT @MaitrixOrg: āļø Long CoT (O1/R1) style reasoning has gained popularity recently. Rather than directly generating the solution, it exhibiā¦.
0
8
0
RT @michiyasunaga: š¢ Introducing Multimodal RewardBench:. A holistic, human-annotated benchmark for evaluating VLM reward models or judgesā¦.
0
37
0
RT @jaseweston: šØ New paper & dataset! šØ.NaturalReasoning: Reasoning in the Wild with 2.8M Challenging Questions.- Synthesizes 2.8M challenā¦.
0
90
0
RT @tydsh: Great to see that many of our previous works on sparse self-attention (StreamingLLM, H2O, MagicPIG) are mentioned in this greatā¦.
0
25
0
RT @tydsh: Our new work Spectral Journey shows a surprising finding: when a 2-layer Transformer is learned to prediā¦.
arxiv.org
Decoder-only transformers lead to a step-change in capability of large language models. However, opinions are mixed as to whether they are really planning or reasoning. A path to making progress...
0
88
0
RT @fly51fly: [LG] Spectral Journey: How Transformers Predict the Shortest Path.A Cohen, A Gromov, K Yang, Y Tian [Meta] (2025). https://t.cā¦.
0
5
0
RT @jaseweston: š„„šŖļø Introducing CoCoMix - a LLM pretraining framework that predicts concepts and mixes them into its hidden state to improvā¦.
0
59
0
RT @qqyuzu: Widely accepted: the longer CoT the better perf - in TEXT space. What happens in LATENT space? We use latent discrete tokens toā¦.
0
17
0
RT @tydsh: Our Coconut work (learning continuous latent CoT) has opened sourced now. Welcome to play with it:
github.com
Training Large Language Model to Reason in a Continuous Latent Space - facebookresearch/coconut
0
270
0
RT @aramHmarkosyan: We're excited to open-source LeanUniverse! A package that simplifies building consistent #Lean4 training datasets fromā¦.
github.com
LeanUniverse: A Library for Consistent and Scalable Lean4 Dataset Management - facebookresearch/LeanUniverse
0
51
0
RT @michiyasunaga: š£ Introducing ALMA: Alignment with Minimal Annotation. Idea: Conventional LLM alignment (post-trā¦.
0
40
0
RT @gh_marjan: Everyoneās talking about synthetic data generation ā but whatās the recipe for scaling it without model collapse? š¤. Meet ALā¦.
0
10
0
RT @Ahmad_Al_Dahle: Introducing Llama 3.3 ā a new 70B model that delivers the performance of our 405B model but is easier & more cost-efficā¦.
0
477
0
RT @ArmenAgha: Say hello to our new company Perceptron AI. Foundation models transformed the digital realm, now itās time for the physicaā¦.
0
57
0
RT @AkshatS07: 1/n Iām excited to share our new venture, Perceptron AI . With the advancements we have made with AI in the digital world ,ā¦.
0
10
0
RT @qqyuzu: Introducing Dualformer: a new model that integrates fast and slow thinking! By learning with randomized reasoning traces, Dualfā¦.
0
28
0