Reading privacy policies is often helpful. They may use pleasant terms like GDPR. If not, they frequently provide clues about how they intend to use your data, allowing you to determine whether they are merely protecting themselves from potential lawsuits or planning to sell your data.

90% of what most people here would do with AI can be done locally with a $300 graphics card. You can go cheaper, but there is a tradeoff between hardware price and productivity loss due to slow responses.

I don't think I'll debate the financial details of UBI here, as others have done so more comprehensively than I could. I believe there are several models that have been calculated and could potentially work. I would have to read about the specifics myself again. It would certainly not work if you told 20% of the population they would be taxed at 100%, as they would likely just vote you out of office, backed by all the other UBI skeptics who see you coming for their income next.

They didn't pay. And they are quite honest about that. The problem is, that laws that allow for scraping only allow to use the result for TDM. If you then just add all of that to the library for employee use, you have to pay for that or get sued. And it would be sane to strictly separate "use for AI training only" and "use for employees" as the waters are already muddy without dual use.

You are free to use deterministic sampling. For many purposes it's even preferred.

Good services allow you to read the reasoning trace and often it is more interesting than the actual answer.
Don't forget, "reasoning" is not to be taken literally. The point is to have the model fill its context first with a lot of relevant information and multiple perspectives before writing an answer. That does not imply that the answer has to use that text, but it likely steers it into a better direction than an answer from nothing. But especially with the tricky questions a few of the "Wait, maybe ... no, but I should also consider ..." babble in reasoning traces can consider interesting points that help the human to find new perspectives on the problem as well.

Quote: "On this record in this posture, the central library copies were retained even when no longer serving as sources for training copies,“hundreds of engineers” could access them to make copies for other uses, and engineers did make other copies."

In plain English, I would read that as: The engineers read the books from the training library for leisure.

No. You may also not download the content. The difference is, that if you're caught downloading you can be charged for $14 in damages, but if you're uploading the companies claim that you're the top of the illegal distribution pyramid scheme and liable for millions in damages. So they rather sue uploaders because that pays better.

The lawsuit alleges that Anthropic stores the books, allowing employees to also read them rather than using them solely for AI training. And downloading books for free to read them is piracy. I am unsure if employees were really permitted to read the books freely, as that would be a very stupid move of Anthropic, but they should dispute this claim if untrue, as it's a key factor in why the judge ordered them to pay for damages.

Storing the books after training might also violate certain copyright exceptions for text and datamining, that require deletion of training data once it is no longer needed. However, Anthropic could probably (honestly) argue they plan to use the data to train future models. The rules about "delete when no longer needed" are unclear about how long you can keep the data around while you're not training the next model yet.

Many don't preinstall vim, but most have some minimal version of vi. Minimal vi is still annoying if you're used to vim, but better than just having nano.

They implemented them basically as a secondy sidebar. They also have a redesign of the sidebar that gets activated when you use vertical tabs and it sucks from an UI design point of view. Even when you disable vertical tabs again it stays the way (until you disable it in about:config) and you have one sidebar with buttons to toggle the different sidebars and then a second sidebar that looks like glued to the first one.

I'd with they would hire some people who know actual UI/UX and not like in "That UI has a nice gradient" but like in "This UI design handbook that tells the patterns that have good reasons why people invented them says we shouldn't make things that flicker into view and out of view when you move the mouse a few pixels." Their current designs get the details wrong that seem unimportant, but make the things annoying when you're actually using them.

The idea of UBI is to feed from money where people have "too much". Currently this means owing companies, using machines to mass produce, or just owning the right stock. In the future this may also include being able to do things with AI to replace humans to some degree. Someone who (on the top of the pyramid) make money "from nothing" can pay for other's missing UBI and it would only redistribute the money that they get from nothing to the people where the money actually comes from, because the "from nothing" is on the back of the people who are working hard because they are not born lucky into a wealthy family who already owns the machinery or whatever is doing the actual work for them.

I still believe that it is hard to get UBI, because the people with much money have obviously no interest in it and it is usually very hard to do things against their interests. Also even people who would profit from it will debate it is unfair when they profit from it less than others, with arguments that can be exploited by the people who would need to pay for it.

But the question was not if capitalism works without such things, but what are the regulations and safety nets for people needed to have a system based on capitalism that still does not sacrifces unlucky people. Some people even argue with UBI when it comes to AI possibly taking jobs.

I like the concept of UBI, but don't think it is necessary because of AI. I am pretty sure the resources that may be freed by using AI can be used to let people do other work and in the end produce better or more products with the same number of employees. Some companies will try to do it without employees and they will see why that doesn't work, so things may be shaken up for some time.

Sorry, I thought with Obama there were changes to that.

Anyway, the point here was, that I think such a lower limit should be provided, even when people fail despite having the same chances. And on top of that, capitalism can work without becoming Darwinism.

That's a really good question for future systems, as I also see that current ones lack good control over the results.

In the end, you have a similar problem like you had with images: An image can convey more meaning than a thousand words. What did people do? They developed Image-to-Image, Controlnets (for things like poses, normal maps, depth, segmentation, etc.), LoRA for styles and characters, and now you can control all the details that won't work with just a text prompt.

I'm not sure about the controls. I think musicians can tell us more about the factors one would want to manipulate, such as BPM, pitch, and so on, which could be conditioned with simple numbers and how one could control others. A few LoRA already exist for music styles, I've seen three for ACE-Step, but there is not much yet. Audio-to-Audio exist as well, but if you can't sing, it might not improve your audio quality much. I haven't tried it yet, though.