From the Hacker News thread
Recently I was quite surprised how good inversion from the OpenAI ada-002 embeddings can be: twitter.com/thesephist/status/169809573…