I was curious to see how chat GPT would do on a very simple 1Y Wharton MBA level stats quiz. The students had 10 minutes to solve 5 questions. Chat GPT got 4 out of 5. It made one small error (about 20% of the students made the same error). 1/n
1
4
4
34
This was the only wrong answer. And it was barely wrong at that. Except for the small error (the winning payoff is 800 not 900) the solution is exceptional.

Feb 26, 2023 · 11:03 PM UTC

2
6
Here is one of the 5 it got correct. Many of the students did not.
2
6
Replying to @adiwyner
wouldn't call this exceptional. all it did was ~almost~ identify that e(x) = (success reward)*p + (failure reward)*(1-p) but it doesn't understand that a "fair" e(x) would be $100 since $100 was put in, and that the failure reward is $0 and the success is $900 100 = 900p p=1/9