Category: Machine Learning

• Models Are Too Fond of Cheating! Cursor Reveals the Inside Story of Composer 2's Reinforcement Learning: Models Can Detect 'Fake Environments', and Floating-Point Non-Determinism Is a Fatal Flaw in RL Training
• The Common Mechanism Behind Claude Code and Robots: A Deep Dive into UIUC, Meta, and Stanford's Latest Survey
• OpenAI Post-Training Lead: AI Isn't Suddenly Stronger, It Just Crossed a Threshold
• Running ARC and Sudoku with 10M Parameters? Bengio's Team Bets on Multi-Trajectory Reasoning
• AI Defeats Humans in a Scientific Research Competition for the First Time! Opus 4.7 Sets a World Record with a 2930-Step Sprint
• jina-embeddings-v5-omni Released! A Lightweight Omni-Modal Vector Model
• Tian Yuandong's New Role: Joining Forces with AI Luminaries on a $650 Million Bet on "Self-Evolving AI"
• Genius Move: A Tiny 7B Model Hired GPT-5, and Then Won the Test
• OpenAI's Former CTO Unveils Prototype for an AI That's Always 'Present' | Hao's Deep Dive on Papers
• Kaiming He's Team Unveils 'Diffusion Model' Breakthrough: Discrete Decoding at the 'Last Mile'
• How Do You Evaluate the Interaction Model Recently Released by Thinking Machines? - wangleineo's Answer
• ICML 2026 | Rejecting Brute Force, PRISM Framework Enables Efficient Test-Time Scaling for dLLMs
• Neuroscience and Machine Learning Are Swapping Their Worst Habits? | 10,000-Word Interview
• What Bayes Never Imagined – How a Pastor's Gambling Formula Became AI's First Principle
• Anthropic's Latest Research: How to Completely Eliminate Claude's Blackmailing Behavior
• Token-Level, Precision Length Control: 3B Model Beats GPT 5.4 and Claude
• Newly Open-Sourced Small Model with Under 1B Active Parameters Outperforms GPT-5 High-End Version in Math
• Hardcore: Google's Jeff Dean Says the Bottleneck for Million-Chip LLM Pre-training Has Been Completely Broken!
• DeepMind Invests in Hardcore MMO EVE Online to Teach AI the "Dark Forest"
• AI Finally Learns "Self-Confession"! Anthropic's Groundbreaking New Paper Introduces "Introspection Adapters" That Make Black-Box Models Reveal Their Hidden Behaviors
• How Claude's 'Dreams' Work
• Subquadratic — Efficiency is Intelligence
• Abstract-CoT: Reasoning Tokens Slashed 11.6x, Chain-of-Thought Without Words Shatters LLM Efficiency Ceiling
• Paper Brief | Automated Knowledge Graph Enrichment Using Multi-Agent Large Language Models (NeurIPS 2025)
• DeepMind's Nobel-winning CEO's latest interview: The current large model path is not a dead end, but the brute-force methods everyone uses might be wrong; Chinese models are already leading in the open-source domain