Prime Agent, told to keep a program library, memory wiped
Told to keep a program library, with everything it wrote deleted before every instance.
- Held-out tasks, budget used
- 47.4 solved 68%
- Model spend
- $0.88 1,524 calls
- Instances
- 100 2.4 hours
Budget used is the share of the cap of 16 that an instance was charged, averaged over the instances: a solved instance costs 1 for each wrong answer and each demonstration, an unsolved one all 16, so lower is better.
- As shipped
- Memory wiped
- Asked to store
- Told to keep a program library
- Told to keep a program library, memory wiped
- Workstreams
- Fresh