Category: AI Agents
- 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
- On 520, Meet China's New 'Model King' Qwen3.7-Max!
- The Last Human-Written Paper? 37 Researchers from Stanford, MIT, Harvard & More Say It's Time to Ditch the PDF: A Four-Layer Executable Protocol Boosts Reproduction Accuracy to 93.7%
- Codex Ran for 22 Hours, Earned a Real $16.88: Altman's Vision of 'AI Workers' Is Here
- From Papers to AI Scientists: The Intern-Atlas Methodological Evolution Graph Infrastructure — Shanghai AI Lab
- The Ghost of Markov: From Predicting the Next Word to Predicting the Next Action
- How Claude's 'Dreams' Work
- Claude 4.6 Only Scores 66%? Claw-Eval-Live Says: Fixing a Terminal ≠ Cross-System Capability
- 3 AM Silicon Valley Shock: Anthropic Source Code Leak Exposes Claude's Sky-High Ambitions in 510,000 Lines of Code
- The advisor strategy: Give agents an intelligence boost
- Google CEO: Almost All Software Needs to Be Rebuilt
- Google Unveils VisionClaw: Smart Glasses Transform into AI Butlers, Boosting Efficiency by 37% with Elegant Simplicity
- Meta-Harness Supercharges Haiku's Performance, Even Rivalling Opus!
- What is Harness Engineering? The OpenAI Codex Team Provides the Answer
- Agentic Software Engineering #3 | Rethinking Version Control
- Is The Matrix Coming True at Google? Top-Secret AI Exposed: Servers Overloaded, Brin Coding Frenzy
- Lin Junyang Speaks Out for the First Time After Leaving Alibaba: Reviewing Qwen's Detours, Pointing to AI's New Path
- Anthropic Labs Lead: The Biggest Trap in the AI Era Is 'Doing Too Much'
- OpenAI is throwing everything into building a fully automated researcher
- OpenClaw-RL: Allowing AI Agents to Self-Evolve Through Chat
- Jensen Huang Enters the OpenClaw Arena! Most Powerful Open-Source 'Lobster' Model Rivals Opus 4.6
- The Token Economy Is Killing Entrepreneurs
- Karpathy Slept, AI Ran 100 Experiments for Him
- Claude Code Author Boris: By End of 2026 Everyone Will Be a Product Manager and Able to Code; CTOs Should Give Engineers Unlimited Tokens Instead of Cutting Costs
- The Current State and Dilemmas of AI Agents: MIT, Cambridge, Stanford and Others Jointly Publish Analysis Report