Home Creators Posts Import Register Favorites Logout

Downloads

Content

This week, Patrons get a bonus story about the upcoming Olympic Esports Games. Because we love you 20% more than the non-patrons.

Comments

Mother pheasant plucker

2 points to consider. 1. Luddites did not hate or fear technology. That's 1800s propaganda. Luddites were a labor movement every bit in the same vein as a modern union. They broke the machines to get fair wages and treatment. 2. AI is very much hype right now. I'd take with a huge grain of salt the claims of it's hypemen. Consider how hyped self driving cars were in 2015 and where they currently are. There is not a clear path to deliver the promises of AI. When that's brought up, the hypemen generally just get louder and angrier. I'm not saying AI isn't or won't be useful. I am saying you should practice skepticism. I've used it while programming and I've stopped using it. It's capabilities aren't where it's being described. Perhaps it will get there, but it also might not. I suggest giving this a read from someone in the field developing machine learning algorithms. https://ludic.mataroa.blog/blog/i-will-fucking-piledrive-you-if-you-mention-ai-again/

Laura Lusby

I’ve only been listening for 45 seconds but BEST OPENING EVER

I’mNotSayingThat

He bravely gave the tip of his ear to save racism everywhere… I’m sure there’ll be more details in the coming days, but so far it sounds like the gunman was the type who’d have shot up a school or a shopping mall if it hadn’t been for this high-profile event in his area. They’ve been desperate to exploit and weaponize this against their political enemies when so far it looks like a textbook case of the exact problem they’ve spent decades fighting *not* to fix. https://www.nbcnews.com/politics/politics-news/trump-rally-shooter-identified-rcna161757

Seigmann

The center-mass joke literally made me do a spit take. :-D

Gmork

Using AI to create a scary-sounding letter or any other kind of first draft of a product is fine, but FFS DO NOT TAKE ADVICE FROM AI. Especially not legal advice. AI does not think or understand or have your best interest in mind. It creates an approximation of what a final product might look like, but has no way to recognize if the substance of that answer/recommendation is true or false.

Gmork

I'm not as hard on AI as you, but they're going way too far in arguing that it can be relied upon for legal advice or any other kinds of answers. Wikipedia was actually a great example. It's great fun for low stakes trivia, still needs to be fact checked re anything substantive.

Gmork

Noah actually did a Citation Needed ep on the Luddites.

Gmork

I'd trust AI to do stuff I could do faster what I could do on my own and then check it. I'd never rely on it for answers that I lack the ability to check.

M R

Respectfully, I feel your Google and AI metaphor isn’t valid. Google’s job is to make the vast amount of information on the internet searchable (well, that was their original job; now it’s to vacuum up all your personal information to sell it to the highest bidder). Google isn’t generally an author of any kind of work, and almost always redirects users to the copy-written work for which they were searching. That is tangibly different than leveraging human-generated, copy-written works to train models with neither the permission, nor compensation, of the copyright holders. I feel this would be a better comparison: when I was in school, I wasn’t allowed to scrape and consume every written work for free. My schools had libraries where digital and physical works were purchased specifically for my classmates’ and my consumption/training/learning. The same went for any subscriptions to publications like the New York Times, the Wall Street Journal, and so on. I don’t think it’s unreasonable to think that what LLM companies like OpenAI are doing here is fundamentally different than what you seem to think constitutes fair use, and leveraging copy-written work without getting the permission of the copyright holders is wrong (as can be evidenced by the agreements OpenAI and other “AI” companies have already reached with many publishers). I’m not sure I would take a bunch of extremely well-funded companies’ attempts to leverage copy-written works without their rights holders’ permission as being such a “cash money” or chill thing as you all seem to.

Sylvie Seeker

100% endorse this, and my Masters degree is in Machine Learning.

I’mNotSayingThat

They desperately want to blame democrats when the shooter was part of a societal malaise that they’ve been roadblocking any and all efforts to alleviate in favor of “thoughts and prayers” since Columbine.

Matt Hiller

Always enjoy when Curling is mentioned. Even if highlighting how unathletic you can be and still play!

Viridian-Flow

Serious question because I haven't heard anyone actually answer it before, but what is fundamentally different from an AI looking at a work and leaning the tropes and style, and a human who does the same thing beyond the AI is significantly more efficient at it? Because like, if people didn't look at styles, figure out what makes that style look like it does and then copy them nobody would know what you meant when you say "anime style" or "noir" etc.. People look at works of art and then they copy it, that's how any artist learns to art. Why is it stealing when AI looks at art and turns out into math but it isn't when a human looks at a Pikachu and tries to draw that?

M R

For me, I feel there are two different conversations going on that often get conflated/mixed up: the training of a model vs. what that model then “creates”. My comment was very intentionally focused on the training side as I feel that is what the crew seemed to be hand-waiving away on this week’s episode. To use your analogy: unless they are pirating content, humans pay for subscriptions to anime streaming services, buy works in the noir genre that grant them the permission of the original author to read their work, etc. What many of these authors are currently objecting to, at least via their lawsuits as I understand them, is the lack of compensation and permission to use their works in training these companies’ models. It’s their stuff, and they are arguing that these LLM companies essentially stole their property as they did not get permission to use it. That argument may not hold up in court, but that’s why one brings lawsuits: to try and prove their case and get the court to act. As for the other side of the ongoing arguments, the one over what the LLMs then “create” once they have been trained, I don’t know that there is anyone out there who can predict what the introduction of generated “works” will mean to us as a society, especially as they get better. You did specifically ask how all this is different than a human doing something similar, and I would say that a human can’t do something similar. I guess we can try to imagine a hypothetical savant who can consume every known human work over a handful of months and then start spitting out thousands of their own “works” every second, but we can see that the analogy breaks down basically from the start. And I think that’s the point: this is completely different from anything a human can do. As the hosts correctly said on this week’s episode: this isn’t fundamentally different from what mechanization and automation have done on the physical side of human creation and labor for millennia now. The biggest (and probably the most scary for many people) departure with the advent of LLMs is that they aim to replace humans doing the stuff that all the other automation has promised to free us up to pursue: creative works. I think most people don’t understand, and therefore fear, what will be left for them if the machines “take away” the value of their physical AND their creative work. And I’m not just talking from an existential perspective, though there is that, but also from a monetary perspective. If human physical and creative labor no longer has monetary value, how do individuals put food on the table, how do they contribute to society and better their lives? I don’t know that anyone has the answers to any of these questions. I just personally objected to the hosts’ seemingly flippantly stating that the large scale consumption of copy-written works without the permission of the rights holders by the products of these LLM companies was somehow chill, even obviously valid. I can kind of understand the publishers’ point and am interested to see how the courts rule.

Bunny Bismuth, Bun-Bun of Seething

We get it, Eli, ChatGPT is your favorite toy. Now why don't you talk to actual writers and artists, or even some computer experts, about the problems with LLMs and other "AI" programs?

Bunny Bismuth, Bun-Bun of Seething

The thing about Wikipedia is that you can actually see where the authors source their data. Can't do that when it's ChatGPT pulling shit out of its virtual asshole!

Jeff Nathan

Big fan, but here to reiterate that Eli significantly misrepresented both the difference between search index and LLMs and how LLMs work. LLMs area are trained on content, some of it copyrighted, and it is the underlying data that dictates in large part how and what the LLM will produce. There are lots of words, but the reason an LLM constructs the sentences it returns are the result of how frequently authors have used words in proximity to each other. Search indexes return results on where to find others work. LLMs create "new" works by constructing sentences based on the probability of data being in proximity to other data (e.g. words in sentences and sentences related to other sentences). These relationships are often the result of copyrighted works and trying to draw a parallel between fair use in search indexing and LLMs seems misguided.