Epoch AI Research described the performance of the agents as underwhelming. While critics of such tests often cite the use of old models or weak baselines, the research team indicated a twist in their findings that addresses those specific concerns.
Tuesday, October 20, 2026
Story
Epoch AI's InnovationEval Finds AI Agents Fail to Automate Research
AI13 days ago
AI agents produced disappointing results when tasked with creating technical breakthroughs in a new test developed by Epoch AI Research. The benchmark, called InnovationEval, measures if AI can generate post-training innovations that match the scale of recent human advances to determine how close the industry is to building an automated AI researcher.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.