ML//model//GPT//GPT-3

GPT-3 is the 175B-parameter GPT whose paper changed the trajectory of the field. It showed few-shot learning without gradient updates: examples shown in the prompt were enough. It was also the proof that scaling laws predict real jumps in capability, beyond a lower perplexity.

GPT-3 is the 175B-parameter GPT whose paper changed the trajectory of the field. It showed few-shot learning without gradient updates: examples shown in the prompt were enough. It was also the proof that scaling laws predict real jumps in capability, beyond a lower perplexity.

When its API launched, anyone could suddenly build on a language model, and prompt engineering was born.