Can LLM agents learn by dreaming? DreamGym from Meta is a new framework that lets AI agents train via synthetic reasoning-based experiences instead of costly real rollouts. It models environment dynamics, replays and adapts tasks, and even improves sim-to-real transfer.
RESEARCH
-

ReasonMed: 370k Medical Reasoning Dataset Advances Clinical AI
By
–
ReasonMed: the largest medical reasoning dataset, advancing LLM performance in clinical QA. Comprising 370k curated examples distilled from 1.75M reasoning paths, ReasonMed is built through a multi-agent EMD (easy–medium–difficult) pipeline with generation, verification, and an
-

Kimi K2 Thinking: Open-Source Reasoning Agent Breakthrough
By
–
Introducing Kimi K2 Thinking — A New Era of Open-Source Thinking Agents! The frontier of autonomous reasoning, intelligent tool use, and scalable cognition has arrived. SOTA Performance: HLE (44.9%) | BrowseComp (60.2%) Handles 200–300 sequential tool calls autonomously
-
Prize Recognizes Foundational AI Work: Data, Compute, Algorithms
By
–
I love that the prize recognizes foundational work in all 3 pillars of AI (data, compute, algorithms), congratulations Fei-Fei!!
-

AI Pioneer Accepts Prize, Emphasizes Collective Responsibility for Impact
By
–
It’s an honor to have received the @QEPrize along with my fellow laureates! But it’s also a responsibility. AI’s impact to humanity is in the hands of all of us.
-
Diffusion Models for Language: Image-Only Input Output Approach
By
–
Side quest: Diffusion model but instead image only input/output but for language modeling.
-
Theory from Experimentation: The Role of Data in Deep Learning Research
By
–
It's closer to the other way around — I arrived at these ideas by doing experimental deep learning research and trying to make sense of my results. Theorizing requires data. And once you have the theory you must test it — which is why I made ARC (an experimental test) and
-
PyTorch Contributor Recognized for Framework Growth
By
–
I hope your skin keeps glowing because of PyTorch 😀
jk jk.
You did more to help PyTorch's growth, and then to help course-correct PyTorch several times than you realize. Thanks for your ongoing service. I hope you continue using it for the next like 30 years, even after we get -
PyTorch Framework Transition: Pleasant Developer Experience
By
–
Great run Soumith! Of all the deep learning framework transitions I've been through where I re-wrote ~all of my code (matlab -> caffe -> numpy -> torch -> pytorch), the PyTorch one was most pleasant and now significantly longest lasting. It hit a jackpot of the time in the
-
Kimi-K2: AI tool for research paper analysis and understanding
By
–
Introducing Kimi-K2 for understanding research papers 🚀
— alphaXiv (@askalphaxiv) 6 novembre 2025
Use Kimi-K2 to understand… the Kimi-K2 paper itself
Highlight any section to ask questions and "@" other papers for quick context, comparisons, and benchmark references pic.twitter.com/4a65tQBFaTIntroducing Kimi-K2 for understanding research papers Use Kimi-K2 to understand… the Kimi-K2 paper itself Highlight any section to ask questions and "@" other papers for quick context, comparisons, and benchmark references
