Selfhosted

I notice that you used the free interface without registration. While a bit of a hassle to register, I saw that behind the registration wall they use longer contexts/bigger models. Without registration it's more prone to abuse so they might condense/truncate responses from chats: by design, the bots have zero memory, so the whole chat history needs to be appended at every question => long chats lead to expensive API calls => free users get history truncated or condensed.

Also, the normal system prompt is something like "be a sycophant and always please the user no matter prompt" and it will lead you do stupid stuff if you ask, so you need to go to the settings and change it to something like:

From now on, stop being agreeable and act like a high-level consultant: blunt and honest. Don’t validate me, don’t soften the truth, and don’t flatter me. Challenge my thinking, question my assumptions, and point out the blind spots I’m ignoring. Be direct, rational, and unfiltered. If my reasoning is weak, dismantle it and show me why; if I’m deluding myself or lying to myself, say it; if I’m avoiding something uncomfortable or wasting time, call it out and explain the opportunity cost. Analyze my situation with objectivity and strategic depth, showing me where I’m making excuses, playing small, or underestimating risk and effort.

For example this question that i had: (all paragraphed, writing from memory, i routinely wipe all the chat histories)

can I bypass gpg verification on fedora when using dnf?

gpt-5-mini from chatgpt free:

it's absolutely possible! just use the --nogpgcheck flag

gpt-5.2 from api with the previous system prompt:

i won't tell you how to do that, it's an incredibly stupid idea, a bad habit and doesn't actually solve the problem

finally, don't blindly trust the results as they might still be incorrect, use that as a hint on how to proceed