ethmarks

Ahh, so you're a meta-splitter.

https://xkcd.com/2518/

I had the same problem and authing on the main page fixed it.

If they're going to block unauthed stylesheet requests, surely they could at least make sure that the CF authenticator shows up on every page.

It's possible that 'adding Linux support' would take the form of just making the anticheat optional.

Maybe playing with the anticheat enabled makes you immune to being reported for cheating (because they can verify down to the kernel level that you aren't), but you can still play without it (but without the immunity from being reported).

Obviously they wouldn't do this in today's market because there's no incentive to do so, but if a significant portion of gamers moved to Linux, offering a Linux solution might become a reasonable choice for game studios.

Yes, but browser sandboxing is an absolute marvel of software design that also cost millions and millions of dollars in developers salaries and CVE bounties to develop. Neither Deno nor yt-dlp have anywhere close to millions of dollars to spend on implementing secure JS sandboxing.

But YouTube is the only one that yt-dlp uses Deno for. No other website on yt-dlp's list has put up enough of a fight to merit an external JS runtime; only YouTube.

From the September announcement:

> The JavaScript runtime requirement will only apply to downloading from YouTube. yt-dlp can still be used without it on the other ~thousand sites it supports

Couldn't you use a Raspberry Pi or a mini PC for this?

Is there any significance to the number 348555896224571969? How was this bug discovered and what was the discoverer doing?

I think that the idea is that if you get enough users on Linux, it seems foolish from the game studio's perspective not to add Linux support to their anticheat.

What non-gaming use cases do you imagine people might use these for?

For normal computer use (reading email, watching videos, doing spreadsheets), there are much cheaper and better options available. If somebody wanted a Steam Machine specifically, it'd be for the GPU.

If you needed a lot of GPU compute (for AI or blockchain or whatever), it'd be cheaper to buy or rent a dedicated server with Nvidia H100s rather than buying dozens of Steam Machines.

So the only potential use cases are those that have a significant but not too significant GPU requirement. The only ones I can think of are gaming (which is the intended use case), video editing, and 3D rendering.

Video editing is less of a concern because neither Adobe Premier nor Final Cut Pro will run on Linux (to my knowledge), so you might as well buy a Mac that runs both of those very efficiently and has decent hardware.

So we're left with 3D rendering. If people want to use Steam Machines to render things in Blender, I say "let them", and I assume that Valve does too.

Laptops have lots of components that the Steam Machine doesn't have. The screen, keyboard, touchpad, cameras, microphones, speakers, battery, et cetera are all fairly small costs, but they add up. Plus using a Linux-based OS instead of Windows automatically knocks around $50 off the price because the price doesn't include the cost of an OEM Windows license.

I don't think the Steam Machine will be priced lower than a PS5 or Xbox (unless Valve is willing to burn money in exchange for market share), but I think that it'll be priced significantly lower than an equivalent-spec laptop (which would be in the $600-800 range based on the fact that the Steam Machine has an "AMD RDNA3 28CUs" GPU, which according to Google is roughly equivalent to an Nvidia RTX 4050, laptops containing which are priced around $600-800).

That's a fair point. Thanks for the detailed response.

I'm curious as to what the scam ads you mention actually are. I use an adblocker most of the time, and most of the adverts that I do see are annoying but fairly innocuous. Furniture, insurance, charter schools, social media apps, shitty mobile games, et cetera. I've seen plenty of slightly scummy adverts, but I can't recall seeing many that are really harmful or blatantly fraudulent. I'm curious to hear what adverts other people are seeing that are so outrageous.

Why is this relevant? To be clear, I'm asking from a place of ignorance. Are you saying that because the video player can have the video decoding happen entirely on the GPU, screen recording software can't pick it up? Couldn't the software just read from the GPU buffer?

Could you elaborate on why? It seems to me that YouTube's implicit contract with the user is "these people paid us to show you this advert", not "we vouch for the integrity and veracity of this advert". I obviously agree that it'd be nice if YouTube would put more effort into screening adverts, but I don't see why they're _obligated_ to. I'm happy to be corrected, though.

I imagine there would be ways around this. I know from personal experience that Kazam screen recorder on Firefox on Ubuntu can record anything and everything, including YouTube as well as DRM content on Disney+ and Prime Video.

I bet that it Google really wanted to it could force Firefox in line, but I imagine that actually preventing screen recording would require compliance at the OS level too, and I don't think that even Google could demand changes like that to Linux. Best they could do is block Linux clients from YouTube, but user agent spoofing or emulation could probably circumvent that.

And even if Google does somehow manage to entirely block screen recording, we can always exploit the analog loophole.

hello!

Kind of a sidenote, but I really hate these page transitions. They're way too slow. Especially because the site has a 114-kilobyte 908-rule inline CSS stylesheet. If you're going to make me download this much CSS, at least make sure it doesn't render your site unusable for nearly a full second even after the page loads.

But they're both framework-agnostic.

This snippet works with any framework and any build step.

  <script src="https://unpkg.com/alpinejs@3.x.x/dist/cdn.min.js" defer></script>
  <div x-data="{ 
    today: new Date().toDateString(),
    randomNum: Math.floor(Math.random() \* 100)
  }">
    <p>Today is <span x-text="today"></span></p>
    <p>Random number: <span x-text="randomNum"></span></p>
  </div>

I'll concede that Alpine.js is harder to understand and more verbose than Marko's syntax, but in order to use Marko you have to commit to the Marko framework. If you're willing to choose a framework solely for its JS-in-HTML capabilities, there are much better choices (like SvelteKit that handles JS-in-HTML wonderfully).

Personally, I'd either just put the content in an ID-ed span and have a script to replace the content. Another, perhaps better, way is to use Alpine.JS which excels at this kind of stuff.

How to you expect those laws to be enforced?

It's impossible to determine with 100% confidence whether or not an image/video was AI generated. If the AI-generated image of Steve Jobs had been copied a bunch on the web, a reverse image search would have turned up lots of sources. Watermarks are imperfect and can be removed. There will always be ambiguity.

So either you're underzealous and if there's ambiguity, you err on the side of treating potentially AI-generated images as real. So now you only catch some deepfakes. This is extra bad because by cracking down on AI-generated content, you condition people to believe any image they see. "If it was AI generated, they would have taken it down by now. It must be real".

The alternative is being overzealous and erring on the side of treating potentially genuine images as AI-generated. Now if a journalist takes a photo of a politician doing something scandalous, the politician can just claim it was AI-generated and have it taken down.

It's a no-win situation. I don't believe that the answer is regulation. It'd be great if we could put the genie back in the bottle, but lots of gen-AI tools are local and open-source, so they will always exist and there's nothing to do be done about it. The best thing is to just treat images and videos with a healthy amount of skepticism.

I do think that people will buy more expensive laptops if they are higher quality. People spend multiple thousands of dollars on laptops, especially if they are powerful.

I own a Windows Surface. I personally find it very convenient, although I'll concede this point if you say the economics aren't working out.

I don't see the problem with creating a token router. Openrouter exists and seems to be doing fine. I'd love if you could actually elaborate on your concerns.

I agree that the CDN idea is pretty weak, but you could make it a bit better by rewarding users based on uptime. Rather than basing the rewards on total time opted in, base it off of total successful requests.

> There is nothing totalitarian about constraining societal harm.

Of course not. The problem is that the only way to enforce AI regulations is through totalitarian means.

You can easily regulate OpenAI and Gemini and whatnot, but then people will just use local AI models. The barrier to entry for using local AI models is basically zero because software like Ollama make it trivially easy to set up, and small LLMs can run on an iphone or a laptop.

The only way to actually prevent "unauthorized" AI usage is to control digital communications to prevent LLM weights from being distributed and to ensure that no "black market" AIs are being trained or used.

And if you're already scanning digital communications for LLM weights, why not also scan it for other forms of wrongthink?. Hey presto, now you have totalitarianism.

I don't think that LLMs fall into the category of an infohazard in the way that Bostrom defines it. It presents a risk, but not one severe enough to justify universal surveillance. Bostrom is talking about AI that can make bioweapons, not ones that gives false confidence to incompetent people.

I think the point is that nobody will give companies money unless the product already works. No more "but in a month this'll get a really cool update that'll add all these features". If you can't trust that a company will continue to exist, you have to be confident that what you're buying is acceptable in its current state.

It's possible that they purchased the movies (although definitely without the proper licensing; buying a DVD allows for personal use, not training a commercial model), or maybe they simply pirated them.

It's also possible that models' entire understanding of the aesthetic comes from screenshots of the movie. Even if OpenAI didn't feed in each frame of the movie, they definitely fed in lots of images from the web, some of which were almost certainly movie screenshots.

That's not relavent. You can't sue someone for benefitting from your work. You can sue someone for intellectual property theft, which is what OpenAI did.

Nobody expects the Linux distribution!

I assume the irony is that an article about signal strength, which is closely related to data speed, is loading slowly because the site servers have slow data speed.

You've convinced me. I did not consider the human cost of producing training data, I did not consider whether or not LLMs were actually saving effort, and I did not consider the extra effort to verify LLM output. I have nothing more to add other than to thank you for taking the time to write such a persuasive and high-quality reply. The internet would be a better place if there were more people like you on it. Thank you for making me less wrong.

That sounds delightful. Get a Raspberry Pi or something connected to a power supply capped at 20 watts (approximate electricity consumption of the human brain). It has to be able to run its algorithm in less than the time limit per turn for speed chess. Then you'd have to choose an algorithm based on if it produces high-quality guesses before arriving at its final answer so that if it runs out of time it can still make a move. I wonder if this is already a thing?

That's a great point, and I think I was being vague before.

To clarify, I was making a broad statement about automation in general. Running an automated loom is more efficient in every way that getting humans to weave cloth by hand. For most tasks, automation is more efficient.

However, there are tasks that humans can still do more efficiently than our current engines of automation. Go is a good example because humans are really good at it and it AlphaGo can only sometimes beat the top players despite massive training and inference costs.

On the other hand, I would dispute that LLMs fall into this category, at least for most tasks, because we have to factor in marginal setup costs too. I think that raising from infancy all of the humans needed to match the output speed of an LLM has a greater cost than training the LLM. Even if you include the cost of mining the metal and powering the factories necessary to build the machines that the LLMs run on. I'm not 100% confident in this statement, but I do think that it's much closer than you seem to think. Supporting the systems that support the systems that support humans takes a lot of resources.

To use your blueberries example, while the cost of keeping the blueberries cold isn't much, growing a single serving of blueberries requires around 95 liters of water[1]. In a similar vein, the efficiency of the human brain is almost irrelevant because the 20 watts of energy consumed by the brain is akin from a resource consumption perspective to the electricity consumed by the monitor to read out the LLM's output: it's the last step in the process, but without the resource-guzzling system behind it, it doesn't work. Just as the monitor doesn't work without the data center which doesn't work without electricity, your brain doesn't work without your body which doesn't work without food which doesn't get produced without water.

As sramam mentioned, these kinds of utilitarian calculations tend to seem pretty inhuman. However, most of the time, the calculations turn out in favor of automation. If they didn't, companies wouldn't be paying for automated systems (this logic doesn't apply to hype-based markets like AI. I'm talking more about markets that are stably automated like textile manufacturing). If you want an anti-automation argument, you'll have a better time arguing based on ethics instead of efficiency.

Again, thanks for the Go example. I genuinely didn't consider the tasks where humans are more efficient than automation.

[1]: https://watercalculator.org/water-footprint-of-food-guide/

I think we're making similar points, but I kind of phrased it weirdly. I agree that current LLMs are sensitive to phrasing and are highly unpredictable and therefore aren't useful in AI-based backends. The point I'm making is that these issues are potentially solvable with better AI and don't philosophically invalidate the idea of a non-programmatic backend.

One could imagine a hypothetical AI model that can do a pretty good job of understanding vague requests, properly refusing irrelevant requests (if you ask a mechanic to bake you a cake he'll likely tell you to go away), and behaving more or less consistently. It is acceptable for an AI-based backend to have a non-zero failure rate. If a mechanic was distracted or misheard you or was just feeling really spiteful, it's not inconceivable that he would replace your engine instead of changing your oil. The critical point is that this happens very, very rarely and 99.99% of the time he will change your oil correctly. Current LLMs have far too high of a failure rate to be useful, but having a failure rate at all is not a non-starter for being useful.