> but as GPT-4 shows, it will take more than scaling to go the remaining 10%. That is happening though; many scientists are working on making gpts that can lea

> but as GPT-4 shows, it will take more than scaling to go the remaining 10%. That is happening though; many scientists are working on making gpts that can learn like Alpha Go Zero where it can learn from scratch with far less information to become better (more ‘intelligent’) quality. Didn’t see anything coming out yet; it is at least one thing that I think could be an AI winter if not achieved before, say 2027. We cannot keep scaling and thinking the results will improve, and, also, indeed we lack training data already, so something must change to need as much; it needs to learn smarter.