Hacker News Reader

Comment by wlesieutre | Hacker News Reader

On the same topic from another webcomic: https://www.penny-arcade.com/comic/2024/01/03/terminarter

(text commentary below the comic, in case your OS has decided to conceal the scroll bar from you and you didn't notice the page is longer)

View on HN

> I'm always reticent to fully engage in "The Dialogue," regardless of its momentary configuration. It's a smoothie made from shibboleths; you have to be able to detect them at only a few parts per million because once these things metastasize, they stop being about whatever they were about and instead become increasingly loud recitations of various catechisms and loyalty oaths.

Boy does that ring true.

Yes, the entire post (it's short) is well worth a read.

And from the following day, their paraphrasing of https://www.theverge.com/2023/11/4/23946353/generative-ai-co... is succinct:

> Meta: We stole so much stuff that, actually, we didn't steal any stuff

> Google: If things were different, things would be totally different

> Andreeson Horowitz: But we already spent so much money

> Microsoft: Think about how this would hurt the little guys, like us

> Anthropic: Fucking shut up about it

> Hugging Face: It's super legal but honestly it might not be idk

> StabilityAI: It's legal in places that aren't here

This is a little oblique, but isn't there a parallel to the music piracy debates of the 90s here? The defense of those who steal training data, "It's not stealing because I'm only taking a little tiny piece from a whole bunch of people", seems like a neat inversion of the defense offered by people torrenting mp3s in their dorm rooms.

It's worth noting that, back when it suddenly became technically feasible to copy all songs for free, the response from the rightsholders of music was, essentially, to demand that we all collectively pretend it isn't. It might make sense for the rightsholders of training data to try to do the same, if they could speak with one voice and if they had lobbyists, but I guess the fact that they can't and don't settles that.

I like a good metaphor, so I see your point, though it doesn't quite fit for me.

To me, training a model is less like redistribution and more like reading something and having it influence your thoughts. You may be able to reproduce certain memorable sections of things you've read verbatim (common with poetry, for example) and you may well be able to reproduce a 'style', but you don't have the entirety of the source material memorized in a way that it could be redistributed in full (but if you did, that may be well an infringement at such a time).

I agree that copying music is not all that similar to an AI model using training data, but they're both examples of a superficially similar activity being transformed by automation and scale. The difference between manually copying a song on cassette for a friend and uploading that song to napster resembles the difference between a human learning to draw in an established artist's style to an AI model learning to draw in an established artist's style. In both cases, the offender's defense is a variation on, "It was okay when humans did this slowly..."

but if you call it virtue signaling, you get categorized into the group that uses that loyalty oath most often, instead of the non-group member that is simply frustrated by the accuracy of it occurring

well yeah, because you'd have to be virtuous to be able to virtue signal, and the ones who bring that up generally aren't.

It's the implication of virtue signalling that it's a facade?

If Steve the megarapist is writing articles about the latest popular injustice, does that make Steve virtuous?

Even a stopped clock is right twice a day, so they say.

Side comment: I respect the author’s right to choose words that ring well to them, but jeez, as a non native english speaker, reading this, i am happy my device has a dictionary function.

As a native English speaker, I agree!

That's Jerry's "thing." Jerry Holkins has always enjoyed how weird English is and what you can do to someone's brain with a few precisely-chosen words aligned in the right order; it's his writing style.

If he applied his approach to Python coding, he'd be one of those folks who isn't happy until their entire program is a single `do()` method that executes by using decorators to synthesize classes and metaclasses out of thin air, that themselves generate the Python code as strings, that itself gets `eval()`'d to execute the intended program.

Which is all fun and game until you want to get some actual shit done, or in this case, communicate...

They're into their second half century doing this. Would they have made it this far without being a little weird? Maybe, but it's hard to A/B something like that. I understand it well enough after following the comic that long.

It's not for you.

(this is a reference to an ancient Penny Arcade bit: https://www.penny-arcade.com/comic/2004/03/24/the-adventures...)

> It's not for you.

> (this is a reference to an ancient Penny Arcade bit: https://www.penny-arcade.com/comic/2004/03/24/the-adventures...)

Thank you for posting this. I remember it, had forgotten it, and re-reading it after completely forgetting about it is almost as good as reading it for the first time. It still holds up 20 years later!

I think the point is that this is an artistic choice. I consider that paragraph to be halfway between poetry and prose. It's not the style of writing that you would want for most forms of communication, but if you think of it as being akin to a bit of poetry, then perhaps you will find it more acceptable.

I for one really appreciate seeing language being used so artistically. (Though if I was reading a technical spec, product description, legal contract, etc... absolutely not.)

I'm not a native speaker and I enjoy his writing and have no problems understanding it. In this specific case it helps to be a bit of a nerd about religions/mythology though :P

I'm curious if Gabe over at penny-arcade would be upset if his comics were used in an art class to train a new generation of artists.

>> If my interlocutor isn't prepared to make any distinctions between a human being and a machine designed to mimic one, I think that I can't meaningfully discuss this with them. I hate that this is the case; it must be seen as a personal failing. I can't do it, though. These models aren't people. They don't know anything. They don't want anything. They don't need anything. I won't privilege it over a person. And I certainly won't privilege its master.

Perhaps if you read Gabe's post you could have saved yourself the trouble of making this comment.

Seems lazy to differentiate between types of intelligence without providing any substance as to why you believe one type is more worthy of learning than another and criticizing others for the same laziness

>Seems lazy to differentiate between types of intelligence without providing any substance as to why you believe one type is more worthy of learning than another and criticizing others for the same laziness

Seems lazy to call a genAI model "intelligence" without providing any substance as to why one should believe that, but OK.

The model isn't an "intelligence" because it's not making a choice about which data you train on and be "inspired by", as people here say.

The humans that operate those models do. It's those humans that are exploiting the artists.

The AI, as many have said here, is just a tool.

Funny how it's "just a tool" when a human uses it to create art, but it's "intelligence" that gets "inspired" by art when a human uses it to create a software product.

> The model isn't an "intelligence" because it's not making a choice […]

There’s no evidence any of us are making choices either. I don’t think LLMs are intelligences in the way most people mean the word (I think that would be vastly overestimating their abilities) but appealing to choice when it’s so poorly understood — as poorly understood as intelligence, even — is not useful.

>There’s no evidence any of us are making choices either.

There is very clear evidence that the humans that operate the computers make the choice on which data the model gets trained on.

The discussion is about those choices.

My point is that there’s no scientific way to demonstrate that one has made a choice vs that outcome being determined (perhaps probabilistically) or being random. We can’t “go back” and do it a different way to demonstrate we’ve actually chosen.

We don’t understand what it means to be intelligent in the way people generally mean, and we don’t understand (scientifically) what it means to choose. So we can’t use choice to usefully define intelligence.

>My point is that there’s no scientific way to demonstrate that one has made a choice

"I'm sorry, your honor, there is no scientific way to demonstrate that I have made a choice to shoot the victim dead. Therefore, I am not guilty!"

The way that an AI gets a data set to be trained on is via a choice made by humans. You can't reason your way out of the fact that a choice is made, that it's not made by the AI, and that it has consequences.

I mean the interesting bit of this conversation is about intelligence not laws. If you want to debate that talk to someone else I guess.

The conversation is about artists being ripped off by people who train AI models on their work without consent (and profit off of that).

But sure, let's pretend the conversation is about intelligence of an algorithm that has no agency and no say in the matter.

Being obtuse on purpose is fun!

> The model isn't an "intelligence" because it's not making a choice about which data you train on and be "inspired by", as people here say.

This is a very poor definition of intelligence and is not convincing of deserving training data

So says you.

Models aren't capable of existing without humans in the loop making the choice of which data they get trained on.

Consequently, those humans and their choices need to be regulated.

Training people and training AI are completely different things, and this tired argument is disingenuous.

This would explain why I find midjourney art so hideous.

I think a better explanation is that the vast majority of users don't try any prompt parameters or different styles and just rely on its default settings

Sorry, this was a joke about the decline of "Gabe's" artistic output. I'd find midjourney's output repulsive no matter what it looked like.

Got it. Though Midjourney can certainly be pushed beyond its fingerprint, so you might not be aware that a given image you've seen was generated by it or not

Let me be clear: it is the knowledge that an image was created with midjourney that repulses me, not the visual appeal of the image itself.

That is unfortunate

It is also not an uncommon opinion, which you should keep in mind if you intend to use AI-generated imagery in your products.

Actually, it's an opinion that's going to dwindle as AI tooling becomes more commonly available. I am not worried about it

Nah. Even the "good" stuff feels cheap. It's the art version of ho-hos and twinkies. Every once in a while I see one and think, "oh, neat, wonder how they got it to do that?", but usually I'm just left slightly queasy and with a bad taste.

Then it's no problem right? Or are you just in a minority.

That is a non-suitor. A lot of people like hohos and Twinkies.

That doesn't mean they arent stolen and that doesn't mean they are fine cuisine.

Price and volume mean it still has impact. We’re already seeing that with online marketplaces and forums filling up with high-volume AI excreta, and a lot of businesses love replacing positions with software so regardless of the relative merits of the output anyone who lives in a society should be thinking about what it’ll look like as millions of jobs disappear.

I've been using Midjourney to generate assets and textures that I Photoshop into new images, such as album covers, banners, and flyers. It has been excellent for speeding up my workflow. My small audience has found a high amount of visual appeal in the results

Reply on news.ycombinator.com