Part III
Acting and checking
A case where each stopping policy was run to the end of the task, so a decision changes the evidence that follows it.
Chapters
- 3.1
Deciding when an agent is finished
On tasks written for this booklet, an agent's first finish met every checked requirement in 87.5% of runs, sending it back on a model's judgment raised that to 90.0%, about what sending every finish back achieves, and telling it which requirement failed raised that to 94.0%.
DraftRecordsconstructed-agent-tasks