Latest stories
PoS Framework Gives LLM Agents Explicit Belief States to Escape 'Belief Trapping'
A new inference-time framework called PoS maintains explicit belief states for long-horizon LLM agents, detecting and recovering f...
On-Policy vs Off-Policy Distillation: Hugging Face Study Finds Rollout Policy Isn't the Main Driver
A controlled strong-to-weak distillation study across Llama3 and Qwen2.5 finds that token-level KL direction shapes task performan...
OneStreamer: A 4B Model That Learns What to Remember and When to Respond in Streaming Video
Hugging Face researchers introduce OneStreamer, a 4B streaming video LLM that jointly learns query-independent evidence recording...
Chatham Financial Cuts Trade Validation from 30 Minutes to Under 4 with OpenAI
Chatham Financial is using OpenAI's Codex and GPT-5.6 to build technology and redesign workflows, reducing trade validation time f...
New Study Finds Post-Training 'Sharpening Tax' Limits LLM Test-Time Scaling
Researchers propose a diagnostic metric called 'Sharpening Tax' showing that RL post-training improves single-shot accuracy but re...
World Observer Decouples Watching from Acting to Keep World Models Consistent
A KAIST CVLab paper introduces World Observer, a framework that pairs an actor-centric view with panoramic observers so objects re...
Hugging Face Paper Proposes 'Rhetorical Robustness' as New Test for AI Peer Reviewers
A new paper introduces Rhetorical Robustness as a joint requirement of stability under content-preserving rewrites and discriminat...
APPL Uses Structural Priors to Bridge Skill Learning and Composition in Robots
A new Hugging Face paper introduces Agent Priors-guided Policy Learning (APPL), which treats each policy's structural prior as par...
Hugging Face Paper Introduces HC-DLM, a Hybrid Diffusion Language Model
A new research paper proposes Hierarchical Continuous Diffusion Language Models (HC-DLM), which combine discrete token generation...
Hugging Face Paper Proposes Adaptive Reward Routing for Joint Audio-Video Diffusion
A new research paper introduces Adaptive Reward Routing, a method that dynamically adjusts where reward-driven updates act and how...
Manus Flex Lets Users Bring Their Own API Keys Into the Workspace
Manus introduced Flex, a feature that lets users connect their own API key directly inside the Manus workspace, alongside an expan...
OpenAI Argues AI's Biggest Impact May Be in Everyday Execution
In a new essay, OpenAI suggests that advanced AI's most significant contribution may lie in the routine work that underpins major...