
Ep 58: Sam Bowman on ChatGPT & Controlling AI
Brave New World -- hosted by Vasant Dhar
The Results of the JatchPT Modeling Experiment
Human participants who used the model were significantly more accurate than people under the same constraints. We also noticed, and this is not in the version of the paper that's out yet but we should have a follow up soon, this clearly gets better as the models get better. But the problem is still very much open. The sort of people we're working with are still, these questions are still hard, they're still answering them with sort of less than 80% accuracy.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.