Hacker News Reader

Comment by Isamu | Hacker News Reader

The part of the AI generating the mashup doesn’t really understand what the signature means, or how it may be interpreted, it’s just a visual part of a New Yorker cartoon. Elsewhere in ChatGPT there is plenty of information about signatures and what they mean, and what plagiarism is.

It’s not organized like a human brain, it shouldn’t be surprising that unusual results occur. They are approximating human intelligence from a different angle. It’s interesting to see the improvements in areas like this that require introspection that isn’t fully wired up yet.

[edit] I should add that a human making a New Yorker cartoon is extremely iterative and introspective. Current generative AI is meant to push it out, and you can do the iteration and introspection yourself.

View on HN

Us: "Hey, AI is using copyrighted works!"

Them: "AI is not a dumb copying machine; it processes the work so it becomes a derived work; it's just like how a human learns things from existing works and produces new, authentic works."

Us: "Hey, AI is copying signatures!"

Them: "You should view AI as a dumb copying machine; AI is not organized like the human brain."

Us: *commits CFAA felony*

Them: Hacking? 10 years prison!

Us: Hey, you just confessed you committed a CFAA felony!

Them: AI is obviously dangerous and uncontrollable, but venture capital is drying up! IPO incoming! Give us all your retirement fund money!

Kinda like how if I break some plates at a restaurant I have to pay for it, but if Tony Soprano breaks them the restaurant pays him.

Protection is a heck of a business model, if you can get away with it.

The issue of whether the AI-generated images are copyright infringement is entirely separate from whether the AI is capable of independently creating derived works.

See for example https://en.wikipedia.org/wiki/VMG_Salsoul_v._Ciccone which decided that it's allowed to sample certain instruments etc. in music production. Could AI be compared to such sampling? Perhaps.

Though, what actually decides isn't really the merits of one's case, but rather who has the money to fight in court. And we all know who'd win in AI companies vs cartoonists, unfortunately.

Might be an example of the Goomba Fallacy [1]. Though that brings the interesting point of how could you ever know whether the same person/group has contradictory beliefs if you aren't seeing the exact same account posting an example of it. It seems like it could be an easy out to just say "Goomba!" whenever hypocrisy is observed.

[1]: https://en.wiktionary.org/wiki/Goomba_fallacy

> It’s not organized like a human brain, it shouldn’t be surprising that unusual results occur. They are approximating human intelligence from a different angle. It’s interesting to see the improvements in areas like this that require introspection that isn’t fully wired up yet.

AI boosters take note: this sort of thing is exactly what skeptics have in mind when they insist that you are nowhere near "AGI" and have not meaningfully passed Turing tests and your claims of goalpost-shifting are fake. You have been aiming at straw goalposts.

https://news.ycombinator.com/item?id=49928566

Show HN: Giving Opus 5.5 a simulated paint canvas

It's well possible already, it's just inefficient.

Also of course you can (and always could) negative-prompt "artist signature". Or have the LLM give it a checkup with edit tools. This is all just a massive skill issue on the part of the people who set up the tooling.

Being able to fix the problems doesn't change that the system fundamentally works differently.

Diffusion works differently. Nobody is asserting that diffusion is AGI or on a path to AGI.

I don't get the allegations of goalpost shifting around Turing tests?

A couple weeks ago someone on this website argued that expecting AGI to be able to learn from experience was moving the goalposts.

Turing discussed learning from experience extensively in his classic thinking machines paper.

Its a huge stretch saying it's even approximating human intelligence. All llms are lossy, recombinant, compressed search indexes. There is no introspection - and what we mostly see is fairly rudimentary heuristic programming loops wrapped around them - which tells you how powerful a tool llms are, but don't mistake the capabilities of the tool for reasoning in any form.

Do they even have the equivalent of a game loop? I'll bump them up a notch when ChatGPT texts me, wanting to continue the conversation we were having yesterday.

Have you got a proof that humans are not lossy, recombinant, compressed search indexes as well? Because that whole claim depends on it and we don't really know that much about that human brains really are.

The burden of proof does not lie there, this is a popular fallacy people love to reflexively respond with.

If you don't know that much about human brains, then how on earth can you make the claim that these possess human intelligence? The burden of proof doesn't lie on the "prove it's NOT as intelligent as a human" side.

Thinking about this for 5 seconds reveals how silly this line of thinking is, sorry, and this exact comment is made 500 times a day here.

> then how on earth can you make the claim that these possess human intelligence?

I'm not making any such claim. It's not some black and white situation where you automatically take the extreme opposing side if you raise an issue with some take.

> The burden of proof doesn't lie on the "prove it's NOT as intelligent as a human" side.

It's on the side making any specific claim.

> It's on the side making any specific claim.

That isn’t how this kind of debate works.

I make a positive claim - “there is a god.”

Counterclaim says “No it doesn’t.”

The second claim cannot be made to produce contrary evidence to the first claim. Extraordinary claims require extraordinary evidence.

I think you're getting into a false dichotomy where true/false are the only options. You claimed LLMs don't approach human intelligence and LLMs are lossy search indexes. I'm saying: Does that comparison even make sense? Is intelligence not possible given large enough search? What is human intelligence? What does it even mean to approach human intelligence? And as a specific counter-example: Can we prove human intelligence is not a large enough search? (Chinese room covers that so it's not like we'll discover something new here)

I'm not claiming something contrary. I'm raising an example for "Is the context of the claim even well formed and valid?"

(Same applies to your example: What do you mean by god? Is "is" even a valid state for a god?)

You say it much better than me. Dumb iteration at superhuman speed all the way down.

Some VC executives also lack introspection.

> lossy, recombinant, compressed search indexes. There is no introspection - and what we mostly see is fairly rudimentary heuristic programming loops wrapped around them

I have some bad news for you about human brains

> They are approximating human intelligence from a different angle

Rather, they are aproximating the output of human intelligence.

The process is entirely mechanistic - involving no intelligence at all.

Approximating the output is the goal function.

Intelligence in the mechanism is the consequence of trying to do that efficiently.

Also let's not forget that everything computable - including human thinking - can theoretically be turned into a finitely-sized lookup table, so there's no difference here without asserting human brains work by magic.

> Approximating the output is the goal function.

> Intelligence in the mechanism is the consequence of trying to do that efficiently.

Nice fantasy.

> let's not forget that everything computable - including human thinking

I must have missed that discovery. Which is odd, since it would have been worldwide headline news for sure.

I have used a simple feedback loop to generate AI art: 1) the initial prompt 2) ask for criticism of the generated image 3) apply those changes and generate another image 4) repeat the criticism / fix cycle again

Isn't that somewhat introspective?

Literally, no. Introspection is an internal self-driven process. And typically both the creating and criticisms draw from the general distribution of information, there’s still “nothing” to do the introspecting.

I think your argument would be stronger if you claimed introspection is only a feature of humans.

> Introspection is an internal self-driven process.

I don't think introspection is necessarily self-driven. If a psychologist tells me to examine my thoughts, and so I do, am I not introspecting?

You can replace psychologist with any form of living or non-living derived stimulus but it's only a trigger. Whether you actually engage with introspection, or just react, is still within your control and up to you to guide.

But it gets into the weeds of whether or not we're operating on a shared understanding of what introspection means.

I think it's the different between a psychologist saying "examine your thoughts", and the psychologist saying "examine your thoughts, and please reach x conclusion"

Sure, my point is not that you can’t wire up introspection, just that it’s not currently an extensive part of image generation. I expect that to change over time as better quality is demanded.

Reply on news.ycombinator.com