@rasbt@karpathy Just realized, @rasbt , your book is probably the closest “fine, I’ll do it myself” version we got of @karpathy ’s promised LLM101n course.
@AravSrinivas I have noticed perplexity’s premium GPT-4o setup hallucinates more and has lower context length (for eg. forgets the context completely in 1-2 followup questions) than barebone GPT-4o.
Thoughts?
any ways to improve it?