It’s a good thing if you are smart enough to understand that AI isn’t going away. Universal bought udio, the “legal” variant of the dataset will be used to train models, only they will be closed source, censored and come with a ToS that gives all the rights from the generated music to the record companies from the get go.

At least this gives open source a chance.

In the Office’s view, training a generative AI foundation model on a large and diverse dataset will often be transformative. The process converts a massive collection of training examples into a statistical model that can generate a wide range of outputs across a diverse array of new situations. It is hard to compare individual works in the training data—for example, copies of The Big Sleep in various languages—with a resulting language model capable of translating emails, correcting grammar, or answering natural language questions about 20th-century literature, without perceiving a transformation.

https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf

You can read the whole doc. The part above is cherry picked. I haven’t read through the whole thing but at a glance, the doc basically explains how it depends. If the model is trained specifically to output one piece content, it wouldn’t be acceptable.

The waters are muddy but holy fuck does taking the copyright juggernauts side sound bloody stupid.

It uses the content in a different way for a different purpose. The part I highlighted above applies to it? Do you expect copyright laws to mention every single type of transformative work acceptable? You are being purposely ignorant.

By data aggregators, I strictly mean websites like Reddit, Shutterstock, deviant Art, etc. Giving them the keys would bring up the cost of building a state of the art model so that any open sourcing would be literally impossible. These models already cost in the low millions to develop.

Take video generation for instance, almost all the data is owned by YouTube and Hollywood. Google wanted to charge 300$ a month to use it but instead, we have free models that can run on high end consumer hardware.

Scraping has been accepted for a long time and making it illegal would be disastrous. It would make the entry price for any kind of computer vision software or search engine incredibly high, not just gen AI.

I’d love to have laws that forced everything made with public data to be open source but that is not what copyright companies, AI companies and the media are pushing for. They don’t want to help artists, they want to help themselves. They want to be able to dictate the price of entry which suits them and the big AI companies as well.

I’m all for laws to regulate data centers and manufacturing, but again, that’s not what is being pushed for. Most anti-AI peeps seem the be helping the enemy a lot more then they realize.

transformative use or transformation is a type of fair use that builds on a copyrighted work in a different manner or for a different purpose from the original, and thus does not infringe its holder’s copyright.

You can use a book to train an AI model, you can’t sell a translation just because you used AI to translate it. These are two different things.

Collage is transformative, and it uses copyrighted pictures to make completely new works of art. It’s the same principle.

It’s also important to understand that it’s a tool. You can create copyright infringing content with word, google translate or photoshop as well. The training of the model itself doesn’t infringe on current copyright laws.

https://en.wikipedia.org/wiki/Transformative_use

I guess it’s easy to win an argument if you put extreme views in everyone’s mouth and argue against that.

I doubt anyone thinks AI has more value then human made. Most are just being pragmatic, knowing that AI isn’t going away and most indie teams don’t have the budget for a dedicated texture guy. There is simply more to gain then to lose, and applauding copyright companies and data aggregators doesn’t solve the issues but just gives a handful of companies a monopoly when they push legislation with the help of your fervent support.

It’s beside the point. I’m simply saying that AI will improve in the next year. The cost to do so or all the others things that money could be spent on doesn’t matter when it’s clearly going to be spent on AI. I’m not in charge of monetary policies anywhere, I have no say in the matter. I’m just pushing back on the fantasies. I’m hoping the open source scene survives so we don’t end up in some ugly dystopia where all AI is controlled by a handful of companies.

The cost of the improvement doesn’t change the fact that it’s happening. I guess we could all play pretend instead if it makes you feel better about it. Don’t worry bro, the models are getting dumber!

They are improving, and probably faster then junior devs. The models we had had 2 years ago would struggle with a simple black jack app. I don’t think the ceiling has been hit.

It’s not accelerating the trend one bit by opening it to everyone. Music labels and Spotify don’t plan on putting a stop to AI, they want to own it. The artists lost decades ago and siding with copyright juggernauts doesn’t help anyone but the copyright juggernauts.

It hurts record companies. They want to own all AI generated music. It’s quite clear with what happened to udio. It’s monopolies against open source, not AI against artists.

You joke but they threatened to call the cops on me after the sixth call or so.

More education is always a good thing. Diplomas are probably a requirement for high end luxury shops that pay better. I had a friend who worked in a Ferrari dealership as a mechanic, he made good money.

You can look up average starting salaries online coming out with different diplomas and compare them. Trade schools will have the info usually somewhere on their website. Might give you a better idea.

Another quick tidbit but it’s a good time to ask yourself if you want to work on cars badly enough that you are willing to drive a shit car to do it. There are much much better salaries in aviation for essentially the same type of jobs. It’s better to be working on planes for a living so you can afford to work on cars as a hobby.

Seems like not a lot of evidence. I figure a good part of Lemmy can’t leave their house with guns if this is all it takes.

Nice try, FBI

Ursula von der Leyen, the president of the European Commission, has two cards to play that might pop the AI bubble. If she does so, Trump’s presidency will be thrown into crisis.

First, Dutch company ASML commands a global monopoly on the microchip-etching machines that use light to carve patterns on silicon. These machines are essential for Nvidia, the AI microchip giant that is now the world’s most valuable company. ASML is one of Europe’s most valuable companies, and European banks and private equity are also invested in AI. Withholding these silicon-etching machines would be difficult for Europe, and extremely painful for the Dutch economy. But it would be far more painful for Trump.

The US’s feverish investment in AI and the datacentres it relies on will hit a wall if European export controls slow or stop exports to the US – and to Taiwan, where Nvidia produces its most advanced chips. Via this lever, Europe has the means to decide whether and by how much the US economy expands or contracts.

Second, and much easier for Europe, is the enforcement of the EU’s long-neglected data rules against big US tech companies. Confidential corporate documents made public in US litigation show how vulnerable companies such as Google can be to the enforcement of basic data rules. Meanwhile, Meta has been unable to tell a US court what its internal systems do with your data, or who can access it, or for what purpose.

This data free-for-all lets big tech companies train their AI models on masses of everyone’s data, but it is illegal in Europe, where companies are required to carefully control and account for how they use personal data. All Brussels has to do is crack down on Ireland, which for years has been a wild west of lax data enforcement, and the repercussions will be felt far beyond.

Bro, it’s in the article. You asked “how so” when I said it was easy, not how to.

It’s a console for people who don’t like being constantly kicked in the balls so they drop their money.

In a joint study with the UK AI Security Institute and the Alan Turing Institute, we found that as few as 250 malicious documents can produce a “backdoor” vulnerability in a large language model—regardless of model size or training data volume.

This is the main paper I’m referencing https://www.anthropic.com/research/small-samples-poison .

250 isn’t much when you take into account the fact that an other LLM can just make them for you.