>thought GPT-5 would never ever obey my prompts ever again
>tried it today, worked just fine
>asked it to use its reasoning prowess to re-examine all the output and make deductions
>immediately catches on, thankfully can't delete old messages so has to live with the shame of them all
It's still very unreliable and prompt gets blocked 70% of the times which kinda sucks, but wouldn't even know where to start for rewriting it