Comment by nickpsecurity

Between books and chats, there must be countless examples of someone promising a positive/negative result and the response changing.

Far as proof, I have lists of what many models used, including GPT3, in the "What Do Models Use?" section here:

https://gethisword.com/tech/exploringai/provingwrongdoing.ht...

For GPT3, the use of Common Crawl, WebText, and books will have conversational tactics like the OP used.