Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Things are about to get worse for generative AI

garymarcus.substack.com

Post reply on HN
Print view View original

401–410 of 769 posts

Re: Things are about to get worse for generative AI

#401

Post by gaganyaan »

sensanaty wrote: post #394
stale2002 wrote: post #379

Earlier quoted context omitted.

> “it’s okay because it’s already happening to a lesser degree somewhere else” isn’t the flex you think it is. It actually is. It shows that we as a society are completely OK with this, and nobody is complaining about a very standard and common thing that all artists do. It shows that the outrage is fake, and people don't actually care about the issue.

Do you really see no difference between someone drawing a piece of fan art and trillion dollar corporations stealing other people's works and reselling it for their own profit with no regards to anyone or anything else? And yes, obviously society cares about many things depending on the scales in question. It's okay if a dude goes onto a lake on his small rowboat and catches a few fish for dinner, it's a completely d…

It's not stealing and they're not reselling anything. That's why it's called Generative AI

Re: Things are about to get worse for generative AI

#402

Post by jcgrillo »

pixl97 wrote: post #372
jcgrillo wrote: post #239

Earlier quoted context omitted.

> LLMs could enable them to make the analysis more targeted and effective. How? I'm not trying to be combative, I genuinely am curious if you have an idea how these things could be usefully applied to that problem. In my experience working in the information security space, approximate techniques (neural nets, etc.) haven't gotten much traction. Deterministic detection rules are how we approach the problem of finding…

I guess my next question is how many needles do you find and how sharp are they? Detection rules would filter out most of the noise, then something like an LLM would do a post filter for intent analysis to rank relative risks for human intelligence to look at.

I suspect this would disincentivize operators to take care in the way they write their detection rules, and the nondeterminism of the LLM would then result in false negatives. So the rate of growth of the needles set would increase, and the analysts would be getting lower quality information mediated by the LLM.

In a world where false negatives--i.e. failing to detect a sharp needle--are the worst possible failure mode, approximations need to be handled with exceeding care.

Re: Things are about to get worse for generative AI

#404

Post by sensanaty »

gaganyaan wrote: post #401
sensanaty wrote: post #394

Earlier quoted context omitted.

Do you really see no difference between someone drawing a piece of fan art and trillion dollar corporations stealing other people's works and reselling it for their own profit with no regards to anyone or anything else? And yes, obviously society cares about many things depending on the scales in question. It's okay if a dude goes onto a lake on his small rowboat and catches a few fish for dinner, it's a completely d…

It's not stealing and they're not reselling anything. That's why it's called Generative AI

By that logic I can torrent movies and distribute them all I'd like as long as I call it "Generative Watching" or something like that.

And OpenAI quite literally sells access to their models, and if those models are pushing out verbatim copyrighted works as has been alleged by the NYT, then they are by definition reselling copyrighted works without permission.

Re: Things are about to get worse for generative AI

#405

Post by TheRoque »

koliber wrote: post #286

The responsibility for ensuring that copyrights were not violated fall on the person publishing the work. Whether they drew something themselves, hired an apprentice artists with no legal training to draw something, took a photograph of something, or used AI to create an image should not matter. Why does anyone assume that ChatGPT or other tools would NOT produce previously-copyrighted content? I can see a naive assu…

So it makes generative AI essentially unusable, because you don't know if the output is plagiarism or not, so you'd just doubt it always and never use it.

Re: Things are about to get worse for generative AI

#406

Post by nradov »

FridgeSeal wrote: post #356
Arainach wrote: post #321

Earlier quoted context omitted.

>This is the bit I don’t get from the “feed everything to machine” LLM-maximalists. Do they think courts don’t take context into account, do they think all actions happen in a vacuum and that they can just skip along and ignore laws at their pleasure An entire generation of unicorn startups believed that (Uber, AirBnB, etc.). We see in the news every day that once you have enough money laws don't apply to you (most t…

> Uber and AirBnB The 2 darling startups that are now facing increasingly less rosy futures? Airbnb in particular is facing enough backlash that I’d be surprised if it lasts terribly much longer. Sure, they get away with it for a while, but not forever. > We see in the news every day that once you have enough money laws don't apply to you I agree with you here, but I think this is a much broader conversation about ca…

AirBnB will get away with it forever. While short term rentals might get banned in a handful of cities, the service now operates worldwide. The stock might be overvalued but if you examine their financials it's simply not plausible to think that failure is imminent.

Re: Things are about to get worse for generative AI

#408

Post by tsumnia »

schneems wrote: post #370
tsumnia wrote: post #354

Earlier quoted context omitted.

When I cover generative AI in my Ethics in AI lecture, one of few soapbox opinions I give is that GenAI is doing essentially what people do - copy others. Picasso has a quote about "Good Artists copy, Great Artists steal", which doesn't mean try to pass Lario and Muigi off as your own, but rather that great artists are able to take aspects from other works (also called 'inspiration') without being caught. My personal…

Rounding up a transaction and taking the leftovers wouldn’t be a crime worthy of the FBI for one transaction but it would be for a million or a billion. Scale matters and impact matters. If you’re making an ethical argument “it’s okay because it’s already happening to a lesser degree somewhere else” isn’t the flex you think it is. If you’re talking ethics, talk about impact. Who does it help the most who does it hurt…

> are you teaching the class or taking it?

I teach it, my background is located in my profile and my research focuses on CS education.

Scale and impact do matter, I wholeheartedly agree. However, I stand by my point that genAI is mirroring how humans learn - repetition of previously observed actions. As part of my dissertation, I argued that humans operate using 'templates', or previously established frameworks / systems. Even in higher cognitive tasks like problem solving, we rely on workflows that we were trained on previously. Soloway referred to problem solving as a mental set of "basic recurring plans" [1] and if you look at the old 1980s Usborne children's books, they required kids to retype code [2]. For creative tasks, depending on the actor's background, Method and Meisner both tell people to draw from previous experiences and observations to develop a character. This behavior is similar in many areas like music, dance, martial arts, cooking, language acquisition, etc.

I am not making an ethical argument that GenAI violating copyright is okay because that's what humans do. I'm arguing that GenAI mirrors how humans learn. We observe a behavior and attempt to recreate that behavior. The difference is that humans can extract a fraction of the behavior and utilize it as part of something larger while GenAI cannot to the degree humans do. I'm sure GenAI would struggle to recreate "Who Framed Roger Rabbit?" because of the two polar different visual elements of the film (cartoon and real life).

In regards to your "If you’re talking ethics, talk about impact" section, its a bit of a loaded question. One side of the conversation could state that GenAI is helping many people that do not have confidence in their creative ability to produce their ideas, while the other could state its making it harder for artists.

Yes, it absolutely is hurting artists and I fully support the recent writer's strike over AI concerns. But I do not believe that diminishes how the mathematical models used in GenAI mirror our own skill acquistion.

[1] https://ieeexplore.ieee.org/document/5010283

[2] https://usborne.com/us/books/computer-and-coding-books

Re: Things are about to get worse for generative AI

#409

Post by danielbln »

beepbooptheory wrote: post #215
fallingknife wrote: post #178

Earlier quoted context omitted.

This is not about copyright. Think about it. Would you ever actually use generative AI to pirate something when you could just torrent it? While there may be an argument that generative AI is infringing copyright, it is not really a very good tool for it. And there is a worldwide piracy industry already causing much more financial damage due to infringement. This is really about replacement. The copyright holders in…

Even if this is right, its a shitty consolation. These llms aren't ever going to be an agent of greater democratic, every-man content creation or whatever, its just going to be the transfer of capital from one type of huge company to another. Not much of a future, even if it feels cool for a bit.

Open models are a thing though, how do those fit into things?

Re: Things are about to get worse for generative AI

#410

Post by dartos »

sensanaty wrote: post #398
dartos wrote: post #392

Earlier quoted context omitted.

I mean… neither did any AI.

Wasn't ChatGPT trained on the entirety of Wikipedia? And probably millions of pieces of scientific literature, and arts, and movies and games and and and... Perhaps the hyperbole of the entire corpus of human knowledge isn't quite technically right, but it's close enough.

You’re also assuming these statical models learn in the same way humans do, which is very likely not true.

Tho tbh I’m not really sure what OPs point was

i don’t think the amount of training data is relevant here.

Post reply on HN