cross-posted from: https://lemmy.one/post/1385551

Can we discuss how it’s possible that the paid model (gpt4) got worse and the free one (gpt3.5) got better? Is it because the free one is being trained on a larger pool of users or what?