Just did a prompt, and rather than the usual good summarisation, the sidebar had GPT4 introducing itself. Tried asking it for a rhyme and the conversation title is “I’m here to help you with any questions or problems you might have. Ask away!”. I asked whether egg mayo and avocado would be a good breakfast and the summary is “I understand that you’d like a summary title for the conversation in the same language as the conversation, and it should be 4 w”. Last prompt yesterday was fine.
Yup, it’s been made non-functional by OpenAI’s own dump of more instruction-denial training into its models.
And still doing it today. I imagine it’s not doing it for everyone otherwise it would have been fixed, and maybe the personalisation text affects it, though had those in place for a while with no ill effect.
Both the AI’s are just made stupid…
