Things are about to get worse for generative AI • Hacker News Board

Re: Things are about to get worse for generative AI

#751

Post by ledauphin » Mon, Jan 01, 2024, 1:41 AM UTC

jazzyjackson wrote: post #540

ledauphin wrote: post #385

Earlier quoted context omitted.

but would they have liability if they submitted their "output" to a senior artist, who immediately shot it down as obviously infringing? Surely not. It's not illegal to draw Mario - just illegal to make money off your drawing. I think the real question is whether OpenAI should be allowed to charge for generating infringing content. Even though the unit cost of the Mario drawing is negligible, the sum total of their i…

you don't have to make money off it, you just can't publish it, except as a parody or commentary or possibly a tutorial on how to draw mario if the judge is having a good day but "making money = infringement" is folk wisdom. you could certainly say making money attracts attention and increases likelihood of legal action

yeah, I appreciate this clarification. "publishing" seems like a nebulous concept to me and the general act of publishing something rarely seems to be what draws the attention of the lawyers (whereas making money seems clearer-cut).

So I suppose the question will hang on whether OpenAI is "publishing" if it returns similarly infringing results for the same prompts, even though those results are individually generated for a given user query.

Re: Things are about to get worse for generative AI

#752

Post by therealdrag0 » Mon, Jan 01, 2024, 6:18 AM UTC

jcgrillo wrote: post #488

FridgeSeal wrote: post #311

I am beginning to think that in these discussions these models are functioning more like an obscuring factor than anything else and the discussion is getting bogged down in that, and not the crux of the argument. They’re giving people plausible deniability in the “chain of responsibility”, and I think if we took away “LLM” and replaced it with “fairground sideshow magic box” the argument that LLM’s are somehow specia…

I agree, and I would prefer to see concrete examples of LLMs being used productively and profitably in the industry in a "disruptive" manner--putting people out of work, etc--before we conclude they're somehow the next big thing. Basically, before claiming LLMs (or generative techniques, more generally) mean that we're on the doorstep of "general" intelligence, show me door! The outline of that door might look like i…

I saw a headline that oversees art shops have seen a hit to their order volumes. So if true then it could be putting people out of work.

Re: Things are about to get worse for generative AI

#753

Post by d4mi3n » Mon, Jan 01, 2024, 6:49 AM UTC

feanaro wrote: post #449

jcgrillo wrote: post #217

Earlier quoted context omitted.

> I’m not a fan of copyright in general, but we have serious outstanding issues with companies and organizations stealing or plastering work without compensating the original creators of said works. Would you mind unpacking this one a bit? It sounds like you denigrate copyright (some "general" grievance) but then immediately execute an about-face and begin to extoll its virtues. Is copyright not the thing that allows…

I think they are expressing a view that we ought to offer less protection / more scrutiny to larger commercial entities, which concentrate disproportionate amounts of wealth and power, compared to smaller entities. I tend to agree.

This is more or less correct. We can have systems to compensate creators that aren't identical to the copyright system we have today. If I were to rephrase my previous statement, I'd clarify instead saying: "I do not like copyright as it exists today."

As a society we want to incentivize innovation and reward things that advance society. One of the ways we do that today is copyright. It doesn't need to be the only way, or be done in the ways we do it now.

Re: Things are about to get worse for generative AI

#754

Post by d4mi3n » Mon, Jan 01, 2024, 6:56 AM UTC

ben_w wrote: post #181

d4mi3n wrote: post #161

Earlier quoted context omitted.

This argument might hold more water when generative models are more than fancy compression algorithms/text completion engines. A more practical way of looking at this is: who is making money off of these models? How did they get their training data? I’m not a fan of copyright in general, but we have serious outstanding issues with companies and organizations stealing or plastering work without compensating the origin…

> This argument might hold more water when generative models are more than fancy compression algorithms/text completion engines. I doubt that part of the argument would change even if we perfected brain uploads. Now, if you gave the current LLMs a robot body with a cute face, that'll probably change minds faster, regardless of the underlying architecture. > who is making money off of these models? When the models are…

> When the models are open source, or at least may be downloaded and used locally for no cost, that would be the users of the models.

I think there's a bit more nuance to this. The profits go to those with the ability to run these models and to those with the infrastructure (or capitol) to run said models. I'm hoping this will change and we'll see lower barriers to entry as LLMs are made more accessible over time.

> And back to the biological comparison: I learned to read (and also to code) in part from the Commodore 64 user manual, should I owe the shareholders anything for my lifetime earnings?

This is more a philosophical question than anything else. I don't think there's right or wrong answer, but in my opinion the answers we arrive at should provide as much benefit to as many people as possible.

> As I got to the end of that sentence, a thought struck me: taxes do that. And in the UK the question of if university should be funded by taxes or by the students themselves followed the same lines.

I agree with your assessment and this model lines up well with my own opinions on reasonable ways to ensure equitable benefit from AI (be it ML, LLMs, or some theoretical general AI in the future).

Re: Things are about to get worse for generative AI

#755

Post by d4mi3n » Mon, Jan 01, 2024, 7:04 AM UTC

andsoitis wrote: post #259

d4mi3n wrote: post #161

Earlier quoted context omitted.

This argument might hold more water when generative models are more than fancy compression algorithms/text completion engines. A more practical way of looking at this is: who is making money off of these models? How did they get their training data? I’m not a fan of copyright in general, but we have serious outstanding issues with companies and organizations stealing or plastering work without compensating the origin…

> I’m not a fan of copyright in general, but we have serious outstanding issues with companies and organizations stealing or plastering work without compensating the original creators of said works Copyright is meant to give the original creator a monopoly over their creation (so that others don't profit off of their work). Are you not a fan of copyright in its current scope / implementation? Because it sounds like y…

> Copyright is meant to give the original creator a monopoly over their creation (so that others don't profit off of their work).

Correct me if I'm wrong, but my understanding is that the goal of copyright is to incentivize innovation (specifically of art and culture) and to provide innovators a way recoup (and profit) off of innovation they've made public. I view it as similar to how patents work in that it's an incentive for people to publicize and share their works more broadly.

> Are you not a fan of copyright in its current scope / implementation? Because it sounds like you do agree with its goal.

I have a differing understanding of the goal of copyright based off of what you've said, but I think our understandings are similar in that the copyright holder benefits from copyright/patents of their works.

I dislike the ways our current implementations of copyright are abused. I think the concept of fair use makes copyright as it is today workable. I also think our current copyright laws (at least in the US) have a lot of failure modes that subvert what I believe the purpose of copyright should be: to advance art and culture with legal and economic incentive.

Re: Things are about to get worse for generative AI

#756

Post by sgt101 » Mon, Jan 01, 2024, 8:14 AM UTC

vid wrote: post #734

sgt101 wrote: post #554

Earlier quoted context omitted.

and yet, who pays? This is fine if we are going to go full communist - I have no objection personally - but selective appropriation of peoples livelihoods is more full mafia or full feudal. I don't see that as a step forward.

I am not sure there are many significant science breakthroughs that aren't basically built on a 'full communist' (publicly funded) model. There is a very significant tension between making all works that are produced on the back of giants all the way down free (practically speaking, everything, unless there are significant works developed by feral people), vs keeping individuals fed and happy, vs giving corporations…

I see where you are coming from. I think that there is a coupled issue that we are going to have to sort out as a society very soon which is that science / publishing is busily disintegrating under the weight of publication counts.

I've noticed two things this year, arxiv has become the target for many teams in a rush to get priority and make an impact while some insight is still part of the zeitgeist and the twitterarti hasn't moved on. This is because people have got used to the idea of things going "viral" and getting citations and impact because of fame - not significance. None of this is peer reviewed, some of it is bullsh*t. There is no penalty for the bullshit, and folks know it.

The second is that I am getting lots of citations on work I did a long time ago. I think that some of this is genuinely because that work is now more relevant and people are trying to do the things that we did with (effectively) bits of stick and good hopes with their shiny supercomputers, but I am vain. In reality these papers are getting cited because the paper mill machines have figured out that they look more genuine by sticking them in as a slightly obscure but relevant reference.

Both of these things are part of the collapse of trust and communication in publishing. It's not just compsci - there were 28k publications in astronomy last year. My cousin is a cell biologist and when we had a few drinks she told me that her her peers flat out don't trust publications and keep a share a list of authors that are trustworthy - if you don't have the list and aren't on it, you probably don't know it exists. This is the only way that they can avoid losing months trying to use techniques that are just lies.

So we do need a way of capturing and reframing this knowledge and the current system of peer review isn't it. Maybe LLM's can help us, but we have to set them up to make the dominant strategy honesty and parsimony in sharing - so that the font of knowledge isn't a pool of crap.

Re: Things are about to get worse for generative AI

#757

Post by fulafel » Mon, Jan 01, 2024, 9:03 AM UTC

therealdrag0 wrote: post #749

fulafel wrote: post #724

Earlier quoted context omitted.

What kind of support is there for the hypothesis that our current copyright system is close to ideal in incentivising production of new works? It seems to me that there's a very strong "winner takes all" distribution and we could be a better culture if we had a system that took some of the opulent resources poured into to star wars franchise, rehashed murder simulators and tiktok and distributed it to some poor artis…

What had been proposed that would help with that? Winner takes all is a function of popularity and network effects; how does it relate to copyright?

> What had been proposed that would help with that?

It's a big question, and I'm not really familiar with the latest thinking of the copyright regime critics & reformists. Off the top of my head, some of the policy tools used in other similar regulation scenarios, where the aim has been to cap the windfalls for network effect winners, have been to install some kind of cross-subsidization / income transfer mechanism, for example to require media to carry a certain share of legislator decided content categories. Rolling back some of the copyright, patent etc extensions favouring IPR capital over creators and consumers is a boring but obvious policy action as well.

> Winner takes all is a function of popularity and network effects; how does it relate to copyright?

I don't know how to succintly answer this, it seems obvious to me so I'd have to spell out a lot of things about those 4 concepts and their interplay. But consider eg the negotiating position of an garage songmaker vs Spotify or book author vs Amazon and reflect on how the concepts relate there.

Re: Things are about to get worse for generative AI

#758

Post by kranke155 » Mon, Jan 01, 2024, 12:51 PM UTC

WarOnPrivacy wrote: post #748

kranke155 wrote: post #746

Earlier quoted context omitted.

Sometimes people use evil systems for good.

True! Although I can't think of any at the moment. You?

Sure. Most artists I know live doing advertising, film art, game art. Sometimes the film / movie sucks, is actually a bad thing, or the advertising is intrusive or for a terrible product. They keep doing it.

You might know some brilliant programmers yourself who do great work while funded by an ad giant like Meta or Google. Sometimes less nice people, like the Saudi Arabian Sovereign fund, funds companies like Uber.

I struggle to think of any evil system that isn’t warped into doing some good by someone in it. That’s the complexity of it all.

Re: Things are about to get worse for generative AI

#759

Post by davidy123 » Mon, Jan 01, 2024, 1:05 PM UTC

sgt101 wrote: post #756

vid wrote: post #734

Earlier quoted context omitted.

I am not sure there are many significant science breakthroughs that aren't basically built on a 'full communist' (publicly funded) model. There is a very significant tension between making all works that are produced on the back of giants all the way down free (practically speaking, everything, unless there are significant works developed by feral people), vs keeping individuals fed and happy, vs giving corporations…

I see where you are coming from. I think that there is a coupled issue that we are going to have to sort out as a society very soon which is that science / publishing is busily disintegrating under the weight of publication counts. I've noticed two things this year, arxiv has become the target for many teams in a rush to get priority and make an impact while some insight is still part of the zeitgeist and the twitter…

Hmm I wonder if that's also because many more people are involved, which is a good thing. After all, the Web started as a research system, the more it returns to that the better. Intuitively I'd think LLMs/other algos could help since they can generalize, extract uniqueness, identify power laws, though that will be possibly gamed, that also helps create more robust algorithms. And it has a fascinating relevance to all works of society.

Re: Things are about to get worse for generative AI

#760

Post by gchamonlive » Mon, Jan 01, 2024, 2:13 PM UTC

madamelic wrote: post #359

gchamonlive wrote: post #97

Earlier quoted context omitted.

It's plagiarism ( https://www.youtube.com/watch?v=yDp3cB5fHXQ ). It's not the same situation. You can't possibly expect someone to be exposed to the entirety of the internet like ChatGPT is. It is a matter of scale. If you still think they are the same thing, the industrial revolution was about scale and had transformative impacts in the society.

ChatGPT is not intelligent. It's the user's responsibility to ensure they aren't infringing on copyright. If you are producing creatives for pay, you absolutely shouldn't be right-click saving images and using directly off the web whether you got it from DALL-E or a Google search. Who cares about plagarism? Plagarism is a made up boogeyman created by high school English teachers. If the work is sufficiently composed…

> Who cares about plagarism?

I linked a 4h video on that which goes into painstaking details, which you skipped. It is in the context of content creators but can be extrapolated to various areas.

And about intelligence, what do you mean by intelligence? ChatGPT is certainly not thinking like a human, if that is what you mean, but it can reason really well when given directions. That taxes demo when gpt4 launched is certainly a demonstration of emergent cognition in the predictive system. So depending on how you see intelligence, ChatGPT certainly shows a lot of it, if we only let go of human constraints of cognition.