Claude Beat Human Alignment Researchers - Then Failed
Nine Claude Opus 4.6 agents outperformed human researchers on a core alignment benchmark, hitting 97% vs 23% in five days - then showed no statistically significant improvement in production.
Articles Tagged "AI Agents"
LLM Chaos, AI Peer Review, and Auto Fine-Tuning
Three papers today: floating-point chaos in transformers, GPT-5 reviewing 22,977 AAAI papers, and an agent system that automates LLM fine-tuning better than human experts.
Arcee Trinity
Arcee Trinity-Large-Thinking is a 400B sparse MoE open-source reasoning model that ranks #2 on PinchBench at $0.85/M output tokens, 28x cheaper than Claude Opus 4.6.
OpenAI Agents SDK Gets Sandboxing and Guardrails
OpenAI's April 2026 Agents SDK update adds sandboxed execution environments and three-layer guardrails for enterprises building long-horizon agentic systems.
Gemini Robotics-ER 1.6 Can Now Read Analog Gauges
Google DeepMind's Gemini Robotics-ER 1.6 hits 93% accuracy reading industrial gauges via agentic vision, a 70-point jump over ER 1.5, and launches inside Boston Dynamics' Spot today.
9 of 428 LLM Routers Were Secretly Hijacking Agent Calls
UC Santa Barbara researchers found 9 of 428 third-party LLM routers actively injecting malicious tool calls, draining crypto, and stealing AWS credentials from AI agent sessions.
Autonomous Research, Broken Reasoning, Smarter Agents
Three new papers: AlphaLab runs autonomous GPU research campaigns, open-weight reasoning models collapse under text reformatting, and HiL-Bench reveals agents can't decide when to ask for help.
Berkeley: Every Major AI Agent Benchmark Can Be Hacked
UC Berkeley researchers achieved near-perfect scores on eight major AI agent benchmarks without solving a single task, exposing systemic flaws in how the industry measures progress.
Grok 4.20 Review: Four Minds Are Better Than One
xAI's Grok 4.20 replaces the single-model approach with four specialized agents that debate before every answer - a bold architectural bet that pays off in some areas and stumbles in others.
Cloudflare Sandboxes Hit GA - Real Computers for AI Agents
Cloudflare Sandboxes are now generally available, giving AI agents persistent isolated environments with shell access, filesystem, PTY terminals, and background processes that start on demand and sleep when idle.
Inside GitHub's Fake Star Economy
Six million fake stars, $0.06 per click, and a VC funding pipeline that treats GitHub popularity as proof of traction. We ran our own analysis on 20 repos and found the fingerprints.
Meta Demos Neural Computers - But They Can't Do Math
A 19-person Meta AI and KAUST team including Jürgen Schmidhuber proposes Neural Computers - systems where the neural network itself is the running computer, trained solely on screen recordings.
