ChatGPT is clearly dumber than a generally educated human to me, though not very (it has a shortish context window which makes it bad for writing very long posts/articles; it's not super creative and needs some prodding in order to write decently entertaining stories; sometimes the code it writes doesn't work; it can't provide actually correct street directions that validate against Google Maps).

I think my main crux on this question is whether GPT-4, especially with a fancy AutoGPT/BabyAGI-style wrapper on top, can readily do a lot of different tasks that a generally educated human could do, especially with access to a very large set of plugins. I don't have access to GPT-4 or the AutoGPT framework myself (and the ChaosGPT incident legitimately scared the bejesus out of me, so I don't even want to), but I'm still quite uncertain about this and would like empirical feedback about it.

@duck_master Sheer versatility is what determines my no on this. GPT-4 is certainly more knowledgeable than the average generally educated human, but its ability to solve complex problems efficiently is lacking imo, especially with OpenAI’s restrictions on its API capabilities.

@ThomasMurphy This is a good comment, but any experiments?

@duck_master Not particularly, just from my experience

We race-conditioned on this 😆

HA