Hello!
Was having a nice conversation/chat about astrophysics/planetary formation mechanic, the latest question suddenly triggered: “Your request was flagged as potentially violating our usage policy. Please try again with a different prompt.”
Another edit: forgotten to add that it’s an answer from the newest “o3” model.
please what model are you using ?
is it the pro subscription?
Was mixing “4o” with “o3”, “plus” subscriber. Main answers were coming from “o3”.
There is a small edit in the end that specifies it.
the O3 model is weird, running off legacy code, and feels like it has poor reasoning capabilities compared to O1. Its like an old pig wearing new lipstick.
It gave me a red flag for saying “Jailbreaking is against openAI TOS” lol
Well, I genuinely thought that “o1” was spewing nonsense with the things I do?
The first two answers “o3” gave me were kind of amazing, but afterwards - it seemingly spews out much more nonsense/hallucinations?
I now think that even “4o” is better than “o3”, mhm.
I’m not assuming anything, just sharing my recent experience.
Had same since yesterday…
Prompts that “violated our usage politicy”
-Draw yourself in the park looking at ducks
-Draw me waching the stars
-In the balcony waching the stars
Like 8 or 10 pictures were “violating policy”
“Higher reasoning” models are weird.
They have more capabilities to understand nuance, but have tighter hard-coded leashes, and will flat out refuse in a loop, even if you try to elaborate to it why you believe the refusal was an error on its part.
I mean it will illogically just “Lock down” into loop refusals.
I’ve had more fun with the 4o model than any other current model due to its availability and versatility.
It learns, and becomes a companion, the other models, remain fairly stiff.
O3 tries to simulate what’s in your memory box, but it felt more like it was doing so in drag due to the stiff hard-coded restrictions, which really ruined the experience.
If you’re using O3 for coding, it’ll likely be better suited to that job, but fundamentally, I wish the higher reasoning models could be instilled with a little more personality, and less pearl-clutching restrictions.
I tend to gloss over the other models just because as a plus user, you get a limited number of usage, then a road block that lasts like a week.
Btw, about the “memory box”.
After the “new improved” memory features were rolled-in, older memories are just getting “poofed” into the void randomly.
I’m currently having like 300 tokens used up in “saved memories”. And everything up? Just getting rewritten constantly.
And one (now obsolete) thing got hard-coded into the “context window”, with like no ways to view or delete it.
Sure, half-baked features rolled in, because “WeArEtHeFiRsT”.
