Hacker News
new
past
comments
ask
show
jobs
points
by
akoboldfrying
6 hours ago
|
comments
by
iainmerrick
2 hours ago
|
[-]
I had a similar thought -- rather than fixed benchmarks, you want dynamically-generated tests, specifically designed to exercise newly-exposed corner cases. So the way forward might be antagonistic benchmarks generated by another LLM.
reply