Because even if it winds up being a bad study, it still evokes a deeper, more important “truth.”

I’m being sarcastic but that’s actually what’s going on here.

No mention of Gemini in their blog post on sge And their AI principles doc says

We acknowledge that large language models (LLMs) like those that power generative AI in Search have the potential to generate responses that seem to reflect opinions or emotions, since they have been trained on language that people use to reflect the human experience. We intentionally trained the models that power SGE to refrain from reflecting a persona. It is not designed to respond in the first person, for example, and we fine-tuned the model to provide objective, neutral responses that are corroborated with web results.

So a custom model.

When you use (read, view, listen to…) copyrighted material you’re subject to the licensing rules, no matter if it’s free (as in beer) or not.

You’ve got that backwards. Copyright protects the owner’s right to distribution. Reading, viewing, listening to a work is never copyright infringement. Which is to say that making it publicly available is the owner exercising their rights.

This means that quoting more than what’s considered fair use is a violation of the license, for instance. In practice a human would not be able to quote exactly a 1000 words document just on the first read but “AI” can, thus infringing one of the licensing clauses.

Only on very specific circumstances, with some particular coaxing, can you get an AI to do this with certain works that are widely quoted throughout its training data. There may be some very small scale copyright violations that occur here but it’s largely a technical hurdle that will be overcome before long (i.e. wholesale regurgitation isn’t an actual goal of AI technology).

Some licensing on copyrighted material is also explicitly forbidding to use the full content by automated systems (once they were web crawlers for search engines)

Again, copyright doesn’t govern how you’re allowed to view a work. robots.txt is not a legally enforceable license. At best, the website owner may be able to restrict access via computer access abuse laws, but not copyright. And it would be completely irrelevant to the question of whether or not AI can train on non-internet data sets like books, movies, etc.

It wasn’t Gemini, but the AI generated suggestions added to the top of Google search. But that AI was specifically trained to regurgitate and reference direct from websites, in an effort to minimize the amount of hallucinated answers.

Point is that accessing a website with an adblocker has never been considered a copyright violation.

a much stronger one would be to simply note all of the works with a Creative Commons “No Derivatives” license in the training data, since it is hard to argue that the model checkpoint isn’t derived from the training data.

Not really. First of all, creative commons strictly loosens the copyright restrictions on a work. The strongest license is actually no explicit license i.e. “All Rights Reserved.” No derivatives is already included under full, default, copyright.

Second, derivative has a pretty strict legal definition. It’s not enough to say that the derived work was created using a protected work, or even that the derived work couldn’t exist without the protected work. Some examples: create a word cloud of your favorite book, analyze the tone of news article to help you trade stocks, produce an image containing the most prominent color in every frame of a movie, or create a search index of the words found on all websites on the internet. All of that is absolutely allowed under even the strictest of copyright protections.

Statistical analysis of copyrighted materials, as in training AI, easily clears that same bar.

We’re not just doing this for the money.

We’re doing it for a shitload of money!

They do, though. They purchase data sets from people with licenses, use open source data sets, and/or scrape publicly available data themselves. Worst case they could download pirated data sets, but that’s copyright infringement committed by the entity distributing the data without the legal authority.

Beyond that, copyright doesn’t protect the work from being used to create something else, as long as you’re not distributing significant portions of it. Movie and book reviewers won that legal battle long ago.

The examples they provided were for very widely distributed stories (i.e. present in the data set many times over). The prompts they used were not provided. How many times they had to prompt was not provided. Their results are very difficult to reproduce, if not impossible, especially on newer models.

I mean, sure, it happens. But it’s not a generalizable problem. You’re not going to get it to regurgitate your Lemmy comment, even if they’ve trained on it. You can’t just go and ask it to write Harry Potter and the goblet of fire for you. It’s not the intended purpose of this technology. I expect it’ll largely be a solved problem in 5-10 years, if not sooner.

I mean you do. All the time. We all do. You’re allowed to use them, you’re just not allowed to copy them. It’s in the name you know, copy right.

I know. But we’re both talking about the same thing. Everyone gets irrelevant and ostensibly novel ads all the time. Cat litter, beauty products, diapers, whatever. They just so happen to have focused their attention on cat litter when they just as easily could have focused on dozens of other products and noticed the same result. And, in truth, it’s unlikely that they are actually novel, just unnoticed before.

I’d be incredibly skeptical of the claim that they’ve never been served a cat litter ad. Everybody gets served ads that are misses. They’re obviously easy to ignore which makes it difficult to recall what they were about. But I have no doubt that they would’ve been served cat-related ads plenty of times before. Cats are, after all, one of the most common pets.

The thing is I don’t think it has anything to offer to bring in people from outside the genre. Some people really enjoy it but you kinda have to already be into that kind of thing (DOTA).

I never said story games are shallow. But if the games you like are ones where you can feel like you’ve experienced all the game and the story has to offer in a single playthrough then they are, by definition, shallow. Even a great movie is worth watching multiple times of its story has any appreciable depth. Video games, even more so since there should be more to the story to experience.

Sounds like what you enjoy are shallow, linear story games. To each their own, of course. Glad you’re happy with what PS5 offers you in that regard. But the industry has a lot more to offer than that.

Guns may not cause the mental health issues that make people turn violent, but they do allow violent people to become mass murderers. Video games do neither.

I’m not sure how anybody can look at the way GTA 5 online was monetized to hell and not seriously question how far they’re going to try to go with GTA 6. I’m fully expecting it to leak into GTA 6’s single player with an intense focus on getting more and more out of mtx.

C is just a work around for B and the fact that the technology has no way to identify and overcome harmful biases in its data set and model. This kind of behind the scenes prompt engineering isn’t even unique to diversifying image output, either. It’s a necessity to creating a product that is usable by the general consumer, at least until the technology evolves enough that it can incorporate those lessons directly into the model.

And so my point is, there’s a boatload of problems that stem from the fact that this is early technology and the solutions to those problems haven’t been fully developed yet. But while we are rightfully not upset that the system doesn’t understand that lettuce doesn’t go on the bottom of a burger, we’re for some reason wildly upset that it tries to give our fantasy quasi-historical figures darker skin.

I mean “taking pictures of people who are smiling” is definitely a bias in our culture. How we collectively choose to record information is part of how we encode human biases.

I get what you’re saying in specific circumstances. Sure, a dataset that is built from a single source doesn’t make its biases universal. But these models were trained on a very wide range of sources. Wide enough to cover much of the data we’ve built a culture around.

This is the thing. There’s an incredible number of inaccuracies in the picture, several of which flat out ignore the request in the prompt, and we laugh it off. But the AI makes his skin a little bit darker? Write the Washington Post! Historical accuracy! Outrage!