Researchers Uncover Hidden Ingredients Behind AI Creativity
28 comments
·July 1, 2025bgwalter
nextaccountic
Humans spend years training 24/7 before they can do anything useful. People train even during their sleep, in their dreams. And on top of that, we transmit culture to other people, which accelerate their training.
And that's with the huge "pre-training" data stored in our genetic code (comprising billions of years and evolution), alongside epigenetic inheritance.
ergonaught
Do you have any grasp of how much stuff your brain ingested to enable you to post this?
No, clearly.
tempodox
“Hidden ingredients” ==> none of them understand how and why any of this works (or not). They could be easily defeated by Harry Potter, because he understands magic!
soulofmischief
https://www.debevoise.com/insights/publications/2025/06/anth...
> This led the court to conclude that the “[a]uthors’ complaint is no different than it would be if they complained that training schoolchildren to write well would result in an explosion of competing works.”
Workaccount2
Don't worry, they're all just stochastic parrots[1]
GaggiX
From the paper: "Level ∞: Pattern matching with a soul (humans)"
What am I even reading ahah
Edit: Okay after reading it a bit, this paper is actually pretty funny
scarmig
Your "ideas" are just regurgitations of things you read off the internet; you have no coherent theory of "creativity" beyond some ineffable reference to the sanctity of the human soul.
acedTrex
So true, its well known that "ideas" came around at the same time as the advent of the modern internet.
MangoToupe
How does one distinguish between what some call "hallucinations" and creativity?
ticulatedspline
Hallucinations are just lies one believes, and there's definitely overlap between creativity and lying but hallucinations tend to not have the conscious component of lying.
categorizing the difference with AI it's much the same as with a person, context. if you ask a person what's the capitol of Florida and they tell you "Pink Elephant, and the capitol building is a literal giant pink elephant with an escalator up it's trunk", my how creative, but it's a lie. But you press them and it seems they genuinely believe it and swear up and down they saw it in a book. Now it's a hallucination, though is it creative if they believe they're just regurgitating the contents of a book? technically yes but the creativity is subconscious.
Now if you asked the same person to make up a fictitious capitol to a fake state and got that answer you'd say it was creative, and not a lie or a hallucination since the context was fiction to begin with even if the source of that creative thought comes from the same place in both instances. If there's no objectively correct answer and not a copy of an exiting known then it's "creativity".
The biggest difference is hallucinations are rare in humans, above we'd probably assume the person was being flippant, or didn't know and was a pathological liar (and not a very good one). We don't associate those motives or capacity to AI though, the AI genuinely seems to think that's right, that the response is coming honestly, thus we categorize all factual errors as hallucinations.
jmsdnns
hallucinations is when we dont like it, creativity is when we do
fusionadvocate
You rather have a hallucinated driver or a creative driver coming your way?
jerf
The article is about image generators. Image generators specifically work by starting with noise and then refining the noise into an image. That's not how driving software works and this is not a relevant point.
null
MangoToupe
I'd rather have someone I can hold liable for their decisions, tbh.
yard2010
Hell, I don't want any AI driver coming my way.
add-sub-mul-div
Temperature settings will not get you to David Lynch.
77pt77
Correct. Increasing the temperature will probably result in something that makes more sense that Lynch's output.
MangoToupe
Yes, because the thing we look for in art is... coherence?
Retr0id
The paper: https://arxiv.org/pdf/2412.20292
empath75
> For example, large language models and other AI systems also appear to display creativity, but they don’t harness locality and equivariance.
"Next token" prediction is (primary) local, in the sense that the early layers are largely concerned with grammatical coherence, not semantics, and if you shifted the text input context window by a few paragraphs, it would adjust the output accordingly.
It's not _mathematically_ the same, but i do think the mechanics are similar.
josefritzishere
You can always spot AI marketing. There is this consistent misuse of words like "creativity" which implies intent. AI does not have intent or self-awareness. AI has no concept of objective reality. The word "hallucinations" has the same problem. With no concept of objective reality there is no understanding of the real and the unreal. To quote a popular article, it's bullshitting. All the LLM and algorithmic refinements only improve it's bullshitting. https://www.psypost.org/scholars-ai-isnt-hallucinating-its-b...
hopelite
I am leery of such a claim not just being attention bias, because although it surely is mostly AI gobbledygook, it all looks just like the marketing gobbledygook of pre-AI, ignoring any obvious AI tells.
I think you may just be noticing sloppy attention to detail, i.e., not proofing, relying on AI that is not quite ready, similar to devs just committing AI slop without review.
I suspect someone is going to train a marketing specialized AI at some point that is focused on that specific type of promotional manipulative language of marketing. But, frankly, I don’t see it being long loved either though, because I see marketing being totally nullified by AI. You don’t need marketing when humans are no longer making decisions/buying.
“Human and AI creativity may not be so different”
I guess they need more funding and grants. A human does not need to ingest the entire Internet in order to plagiarize what was read. A human does not need a prompt in order to take action. Two humans can have a conversation that does not collapse immediately.
These people apparently need coaching on the most basic activities. How to solve this in the future? Perhaps women should refuse to procreate with "AI" researchers, who prefer machines anyway.