New benchmark reveals how AI agents learn from long-term tasks

Original: Auto-research is getting a lot of attention, but what we should actually use it for and how to evaluate it are still very good questions.

Source: x.com ↗