TTHE changes the harness of an agent during tests
The method uses no weight update and no answer label. It does improve the baseline harness on 6 benchmarks.
Sources
- TTHE: Test-Time Harness Evolutionarxiv.org
Posted
What matters in AI.
SubscribeThe method uses no weight update and no answer label. It does improve the baseline harness on 6 benchmarks.
Posted
