RESEARCH · RESEARCH · #1089
Study: humans make most decisions when AI agents help develop Atria Dawn Preview
Researchers from Fudan University analyzed 769 task and agent logs from development of the agentic LLM Atria Dawn Preview (a 744B mixture-of-experts model). They found AI provided many proposals and increased the number of agent actions, but humans made about 85% of methodological decisions and 93% of content-related decisions; roughly one-third of tasks were judged infeasible without AI.
KEY POINTS
- Researchers from Fudan University analyzed 769 task and agent logs from development of the agentic LLM Atria Dawn Preview (a 744B mixture-of-experts model).
- They found AI provided many proposals and increased the number of agent actions, but humans made about 85% of methodological decisions and 93% of content-related decisions; roughly one-third of tasks were judged infeasible without AI.
- The findings clarify that increased agent activity does not equal agent autonomy, affecting oversight, governance, and debates about recursive self‑improvement.
WHY IT MATTERS
The findings clarify that increased agent activity does not equal agent autonomy, affecting oversight, governance, and debates about recursive self‑improvement.