mighty obsidianReported by @cursive schooner
#Instructions DON'T WORK
cursive schooner
Starts when I ask it about how many years of experience I have designing games, mods, etc. It makes up answers, I call it out, it lies, pattern repeats.
https://chatgpt.com/g/g-p-675d2128a4b88191b7d2d887124e4953-job-hunting/c/68bb34e8-30a8-8327-b48e-f8ef02c77a88
dusky oyster
Same for me.
cursive schooner
Oh, look! It appears that this problem has been bad enough to infuriate me on past occasions as well. You know, I don't normally bother reporting bugs. I just move on. If it's bad enough I've come here, TWICE... Jesus, what hot garbage this thing is.
Previously reported bug: https://discord.com/channels/974519864045756446/1404009114790789203
dusky oyster
It's GPT-5, don't be surprised, it's just a scam. Own interest of OpenAI. Extrem disgusting business. They actually should never have released GPT-5 if they knew what they were doing, but they didn't.
sudden meteor
It seems we can't share a link to a discussion in a project. That doesn't matter.
Dude, you told the model to do something that it's not supposed to do (modify it's responses in a way that it needed to override your instructions) and then you complain when it doesn't work?
And you openly publish the text that confirms that you told it to violate its protocols, so it lied right back to you, and you're blaming the model? Seriously?
The definition of "context" means we have all of the information to form a response. You didn't provide that here. We would need to see the full discussion, custom instructions for the project, files containing instructions and custom instructions for the user, and probably memories.
That never happens in this forum with these kind of rage posts, because somewhere in there is the obvious reason for the manufactured problem. With full context it will be obvious immediately (already is, you posted it) that the user has tried to abuse the platform and the bot won't let them. So all we see here is what the bot responded, never what the bot was told.
"I'm not going to share my valuable instructions or memories."
Then that's where this dies. Without full context, only people with prejudice and minimal critical thinking skills will put credence in this report as-stated.
You've already reported your concerns to OpenAI with downvotes to responses. I'm guessing anyone who looks at that will shrug and figure you got what you asked for. The system works as designed. It's not a flaw that the system didn't follow your mischievous instructions, it's a feature.
cursive schooner
It seems we can't share a link to a discussion in a project. That doesn't matter...
Dude, go away. This is for BUG REPORTS. I post this here to the developers know. NOT so some volunteer corporate white knight shill can come in and tell me "you're doing it wrong." Take that garbage elsewhere.
I don't even know what the hell you're talking about either. "I'm not going to share my valuable instructions or memories," is not something I ever said. So you're just in here, shilling, making up strawmen so you can beat them doesn't and look shiny in your tin armor because "I set the nasty user right." Please, just stop responding to things I write. Again, this is bug reports, not help requests. Just because YOU don't think there's a problem, doesn't mean there isn't a problem, and coming in to MY bug report and telling me how I'M wrong, doesn't mean 1) that I'm wrong, or 2) that there's no problem. You don't know what you're talking about, so go annoy of someone else.
sudden meteor
I'm laughing and I totally hear ya. Let's step back for a sec.
I'm retired from the guide thing. Neither I nor anyone who does/did the volunteer thing here is a shill. You can stop assuming motivations.
Good luck in getting your concerns addressed.