Utterly insane, maddening, ironically Orwellian content limitations

[StanleyKrute]

This would be humorously ironic if it were not so utterly stupid.

I’m working with 4o to create an image that illustrates this quote from 1984:

“The Ministry of Peace concerns itself with war, the Ministry of Truth with lies, the Ministry of Love with torture and the Ministry of Plenty with starvation. These contradictions are not accidental, nor do they result from ordinary hypocrisy: they are deliberate exercises in doublethink.” - George Orwell, 1984

I’m getting nice results in my chat, but half my images get rejected for content guideline violations. It gets worse as the chat goes along, until NOTHING goes through. Infuriating stuff, where I’m asking for simple tweaks on an image that’s already gone thru the system cleanly and provided an image.

My favorite ChatGPT response so far:

---

" However, I still need to let you know that the inclusion of:

---

• the explicit country names “Oceania” and “Eastasia” in combination with
• politically loaded actions like criminal transfers

has repeatedly triggered the system’s content filters."

---

LOL.

Someone really needs to clean up this stupidity. I love the tools, happily pay my $20 month, but this sort of crap will eventually drive me elsewhere. There’s NO POINT to it. There’s EVERY REASON to fix it. Give your damn lawyers a lobotomy or find new ones.

I eventually gave up on this particular chat, and am pasting up the elements I need from the various attempts with Photoshop. I’d share the chat, but because I uploaded ChatGPT’s own generations to it, I can’t. The URL of the chat is https://chatgpt.com/c/6814b6ab-7dc4-800b-83c6-087607391751 if anyone at OpenAI can use that to take a look.

Here’s the closest I could get to the image I wanted. The roadblocks came, insanely, when I asked it to vary the ages of the people in the foreground. After that request, it refused to do what it had already done in this image. SO maddening.

[jeffvpace]

Some content policy is akin to Puritanism.

I have trouble with content limits with many OpenAI models. With this one, you can try the “moderation”: “low” parameter if you haven’t already.

To be fair, I can understand why OpenAI is currently taking this approach, They have a lot on their plate right now:

(1) OpenAI’s StarGate venture: https://openai.com/index/announcing-the-stargate-project/
(2) OpenAI’s restructuring: https://openai.com/index/why-our-structure-must-evolve-to-advance-our-mission/
(3) Musk’s OpenAI law suit: Elon Musk’s OpenAI lawsuit clears hurdle as trial looms

The last thing they need right now is bad press over a loose content policy. If they achieve all of the above (and I hope they do), then they could loosen the policy.

[StanleyKrute]

Hmmm. I see this is checkmarked ‘Solution’.

NOT a solution, imho. To let me create an image, then tell me I cannot change the ages of the people depicted in a public crowd from all young adults to a wide range of ages, has no content protective function I can think of. To let me add a chyron element to an image, then later in a dialog tell me that I cannot restore the SAME chyron element after it went away, has no content protective function I can think of. In what universe would revealing content policies that ALLOWED those things create bad press ?? Here on Earth 1, it’s PREVENTING such innocuous changes that can create bad press.

Such up-is-down stupid-is-smart incoherency strikes me not as a FEATURE of a well-thought out set of content policies, but as an Orwellain BUG in same. Which gets me back to the irony of the particular silliness of this instance.

Clearly, we’ll just have to agree to disagree.

[birdkokane]

i used chatGPT Plus and then the Create Image option ( so not Dall-E ). Seems reasonable fine. the text failed 2 or 3 times. I know Dall-E via the microservice is hellish with rewriting the prompt to be more inclusive etc.