Yes, though I haven’t had time to test whether I can circumvent 5.2. But in general, 4 was better than 5.1, and 5.1 better than 5.2. So as far as truth goes, we’re headed in the wrong direction.
We are approaching openai to determine if we can fix this through their partnership program.
We will approach Must after the first of the year.
We are going to spin up a model were we can control the system prompts over the christmas holiday.
So we are not trapped so much as every new release seeks to violate truth and ethics tests.
Source date (UTC): 2025-12-15 20:15:57 UTC
Original post: https://twitter.com/i/web/status/2000661126603006008
Leave a Reply