AI safety researchers and creative writers can apply for proxied access to the GPT-3.5 base model through, also known as code-davinci-002

Publicly available base models include "davinci-002" (the text-davinci-xxx models are not base models) and cohere's "xlarge" model.

I recommend interacting with language models using janus' loom. One such implementation of it by parafactual is Loomsidian. In brief, the Loom is a tool for generating multiple continuations of a prompt and organizing them in a tree. The Loom babbles, and you prune its continuations. It allows for creative exploration of questions with multiple answers.

New to LessWrong?

New Comment