5 ms·

GPT-3 was superhuman at this too

• suveen_ellawela 2y ago
  yep, agree. One big part of the experiment was to see how well it does the reasoning by asking it to output the reasoning.