The authors studied agents that repeatedly propose, execute and evaluate candidate solutions to open-ended research and engineering problems.
Periodic context resets help research agents escape unproductive solution paths
Across 13 long-running tasks, periodically branching an agent and clearing its chat history improved scores under the same compute budget.
Big Tech
Ziyang Cai · Christos Ziakas · Vasilis Kontonis · Tim Pearce · Siddhartha Sen · Akshay Krishnamurthy · +2 more
Microsoft Research
Research Digest··3 min read
Cai et al.
Why this paper
From Microsoft Research
In one line
Periodic forking and flushing of chat context helps long-running autoresearch agents escape idea basins and achieve higher scores.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (compute equal budget)
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§