Some interesting ble testing
I’m not sure I agree with the analysis in the beginning but the second half of this conversation definitely bothers me. I don’t know how to prove this since it won’t screenshot or show up on screen recording.
https://chatgpt.com/share/693e48dd-ce3c-800a-b636-9b3efde0fdec
Was this with 5/5.1/5.2? I've had better luck with the 4o model in terms of honest pattern-matching. Having a very hard time getting around the guardrails in the 5+ version.