2930 steps vs 2990 steps, does Opus this round count as true autonomous learning or advanced patchwork?

AIMPACT News, May 15 (UTC+8), according to Beating Monitoring, Prime Intellect announced a two-week autonomous AI research experiment. The research team had Codex (gpt 5.5 xhigh) and Claude Code (opus 4.7 xhigh) independently iterate optimizer solutions in the nanoGPT speed race, attempting to reach the target validation loss in the fewest steps. After approximately 10k experiments and 14k hours of H200 computing power, Opus ultimately broke the human record of 2,990 steps with 2,930 steps. The experiment revealed the current capabilities and limits of AI agents. In the test branch that required the development of new algorithms, both models were unable to run without relying on existing code or papers from the human community.