AIThe Decoderabout 2 hours ago
AI Agents Fail to Innovate in Research
5 min read
Epoch AI and Anthropic find that AI agents like GPT-5.6 Sol and Claude Fable 5 cannot independently invent new research methods. While GPT-5.6 Sol achieved about 35 percent improvement with generous grading, it mostly recycled known techniques. The study concludes humans must fully review AI-generated research because models lack scientific judgment and confidence.
Level
Hype check
Tap to vote and see what everyone thinks.
