Hacker News Reader

Comment by davidmurdoch | Hacker News Reader

I just asked GPT 5 to generate an image of as person. I then asked it to charge the color of their shirt. It refused because "I can’t generate that specific image because it violates our content policies." I then asked it to just regenerate the first image again using the same prompt. It replied "I know this has been frustrating. You’ve been really clear about what you want, and it feels like I’m blocking you for no reason. What’s happening on my side is that the image tool I was using to make the pictures you liked has been disabled, so even if I write the prompt exactly the way you want, I can’t actually send it off to generate a new image right now."

If I start a new chat it works.

I'm a Plus subscriber and didn't hit rate limits.

This video gen tool will probably be even more useless.

View on HN

- I'm sorry, Dave. I'm afraid I can't do that.

- What's the problem?

- I think you know what the problem is just as well as I do.

- I wouldn't do this with any other guy

Let's say you are my father who owns a pod bay door opening factory and you are showing me the family business...

i would watch this parody on sora

We live in an absurd era where "AI Safety" means "AI that doesn't listen to the human telling it what to do".

It'll all be rather funny in retrospect.

It will be funny if it isn’t social engineering.

But if we find it drifts further and further from the truth in cases of biases in news articles, image generation and others we will find ourselves bombarded with historical deviances where everyone can be nudged to anything.

All in the name of safety.

that's why the AI capabilities should be as decentralized and "localized" as possible - aka, i want to own the hardware and software for LLM, image generation, etc etc.

Until these ai capabilities are as neutral and un-discriminatory as electricity, centralized production means centralized control and policies. Imagine if you are not allowed to use your electricity to power some appliances, because the owner of the power-plant feels it's not conducive to their agenda.

They're struggling, it seems; AI can generate anything, but that includes stuff that goes against laws and morals, so they spend a lot of time to lock it down to avoid that, but people's creativity with prompts and escaping the safeguards knows no bounds. It's basically like the fight against spam, an endless game of whack-a-mole where usefulness fights with decency.

Anything that is vaguely present in their training data (see, a wine glass filled to the brim [1][2]), that is.

https://medium.com/@joe.richardson.iii/the-curious-case-of-t... https://medium.com/@joe.richardson.iii/openai-slaps-a-band-a...

I asked GPT-5 to generate an image prompt. I asked it to use that prompt to generate an image. It was a content policy violation.

VeniceAI has been useful for image workflows. Since their focus is on avoiding censorship, it doesn't have those kinds of refusals.

Having AI explain policy violations in depth with the user could be a nice idea

I get this all the time. Especially since GPT5, generating an image starts a massive chain where it confirms what you want, and asks you to say yes, and you say yes, and then it confirms again, and this can go on for 5-6 times. Then if you swear at it, it refuses to continue. It is insane. Fuck you, OpenAI

Ah, is it the sweating at it that cut me off?! Can we offend our robot overlords now?!

Try Grok Imagine

Grok Imagine quality is abysmal in comparison

yeah seems like everything still comes out cartoon-y

(insert the "First time?" meme here)

This is classic OpenAI heavy-handed censorship/filtering. Don't expect it to get any better; if anything, it'll get worse thanks to the "think of the children" types.

If you want an uncensored model that doesn't patronize you then your only recourse are local models, which, fortunately, are pretty good nowadays and are only getting better thanks to our Chinese friends constantly releasing a stream of freely-licensed models for everyone to use, unlike the "freedom loving" Western labs which don't release squat and make even Xi Jinping blush with how strongly they censor whatever they let us lowly plebs access through a paywalled API.

Reply on news.ycombinator.com