Remix.run Logo
▲ throw310822 13 minutes ago

They both depend on how much generalisation you can get behind it. Next token prediction can go from spitting out the most common token after the previous one, to a deep understanding of the text that precedes, from syntax to semantics.