
StrategyQA and Big Bench
Data Skeptic
00:00
Strategy QA Is All Yes or No Questions, Right?
So strategy QA is all yes or no questions, and I believe you said they're balanced. To put that 70% accuracy that the leaderboard is getting to now into sort of a frame of reference, would 5050 be or 50% be my accuracy if I was random guessing? Yeah. So what we found basically is that, okay, so these questions are even challenging for people, right, for humans. If I would give you a bunch of strategy questions, you are likely to fail on some part of them. But if I would let you answer these questions while having access to some web search and join, then you would do well,. And we'll report this number in the
Transcript
Play full episode