> It's clear to anyone from day one how useful those models will be I agree it's clear on day one that these models will be useful. What's less clear is how u

> It's clear to anyone from day one how useful those models will be I agree it's clear on day one that these models will be useful. What's less clear is how useful they are -- a year ago most experts would have told you that an architecture that "predicts the next word" won't be powerful enough to make a model that could lead to AGI-alike abilities (and many people here still argue the same today), but here we are now. > Maybe Mark Zuckerberg was just unlucky to decide on a company pivot two years too early That's just a lack of vision. GPT-2 was released in 2019, GPT-3 in 2020. He had plenty of time to commit to AI development instead of Metaverse.