Published:
TLDR: This post follows the thread of papers authored by Alec Radford that ultimately led to GPT-4. It observes that original motivation for the next-token prediction was as a representation learning mechanism, and there appears to be a gradual (and somewhat accidental) realization that these models could be used for much more…
Published:
TLDR: This post follows the thread of papers authored by Alec Radford that ultimately led to GPT-4. It observes that original motivation for the next-token prediction was as a representation learning mechanism, and there appears to be a gradual (and somewhat accidental) realization that these models could be used for much more…
Published:
TLDR: Making sense of where we are in AI research by looking at the bottlenecks of each machine learning era so far, and where this suggests we’re headed.