I don't think training a model with copyrighted data is any kind of gray zone at all.

Running inference against such a model _might_ be infringing.

But I don't see the legal argument that model training would be.

Think of model training as lossy data compression. You provide an input, it produces a compressed artifact, and it can no longer reproduce the original when you uncompress it. It can produce _something_. It cannot reproduce the original.

On slashdot, we all agree that format shifting copyrighted works should be allowed and unrestricted, right?

If there is applicable case law here about how lossy a "reproduction" can be before copyright has something to say about it, that would be relevant. Do you know of any?

I think you're repeating something that feels emotionally valid but in my experience, isn't factually (or legally) accurate.

I have found AI models to be frustratingly _obedient_ of copyright.

I'll be in a situation where I ask a specific question about what a document says, and the AI repeatedly says the work is copyrighted and it can't just tell me what it says _verbatim_, but it can answer specific questions for me in its own words (while telling me to check the original).

Two examples from my life are procedural questions about what Roberts Rules of Order would require in a given situation, and, what ASD STE-100 says about how to name/word things.

The applicable US law would appear to be USC 17 Sec 106 and 107, which on one hand limits derivative works, but on the other, allows fair use for purposes of teaching, scholarship, or research.

(USC references provided by AI, btw. Which I haven't read)

I used to agree with you, and then oddly enough, about 25-30 years ago, I came across this signature on another regular slashdot posters postings:

"The law, in its majestic equality, forbids the rich as well as the poor to sleep under bridges, to beg in the streets, and to steal bread."

(AI tells me the attribution is "The Red Lily", Anatole France, 1894)

Good points, but to be fair, ultrasonic is currently used to break up kidney stones because it is safe to use it to do so.

It hasn't been "autocomplete" in a long time. Sure, there's a training step based on a corpus of Human language, and the autoregressive process outputs a single token at a time, but reinforcement learning trains specific behaviors beyond merely completing a sentence.

Besides, the best way to write something indistinguishable from what a Human might write is to, well, "think" like a Human.

I've been programming now for over fifty years. I'll also say that I'm using vibe coding for personal projects. So far, I'm getting good results. Here are some tips:

1. Know what you're doing. Right now, I'm working on an interpreter for a DSL. But I already know the basic structure of interpreters. I can tell when the code the AI is turning out is good or bad. When it's bad, I tell it what's good and ask it to correct the bad code. This usually works.

2. Vibe code in small chunks. I didn't say code the interpreter and here's the syntax. I coded it piece by piece - language primitives, environment structure, control flow primitives, etc. Each step of the way I checked the output to make sure it was what I wanted.

3. Let it write tests. Again, AI's are pretty good at looking at a piece of well-written code and building tests for it. If you point out corner cases you want tested, it will crank those out pretty well.

4. Show it what you want. If you want a particular style in the code, let it know. It will follow that extremely well.

5. Use it for boilerplate. Let's say I have five primitive types in my language. I work with the AI to generate the first one. Then I say "Do the same thing for these other four primitives". AI works well off examples, so give it some.

AI works pretty well if you know what you're doing and how to use it. I think of it as a relatively diligent junior coder. And it works about that well.

I think the only Premium TVs left are the business TVs that give you meaningful mechanisms to not have intrusive "Smart" features.

Is there a meaningful difference between a Sony TV that harvests data and won't let you opt-out of "smart" features, and a Wal-mart TV that harvests data and won't let you opt-out of "smart" features?

I guess I am blessed to not be an audiophile and not have flawless supervision :)

FWIW, I have:
- a 20 yo 720p dumb 42" plasma
- a 20 yo 1080P dumb 50" plasma
- a 1yo 4k Samsung 65" TheFrame TV

That last one was a splurge I wanted because the "Art Mode" is just too beautiful, and at the time, Samsung really had the only coherent offering. (I guess there are now "off brand" ArtTV attempts from HiSense and others.. i have no experience with them.)

On the ArtTV, we watch youtube or DVDs or XBox on it a little of the time, and all that stuff looks fine to me on the 65" Samsung. But the TV is otherwise displaying pretty artwork almost all of the time, and whatever Samsung has done with the screen, dimming control, bezel, etc, really does work and really is lovely. And you don't need a service or an app to get the experience - just stick a USB full of public domain masterpieces into the TV.

Even so, the Samsung ecosystem is pretty annoying. I can have it show my images in ArtMode, but i cannot have the "real" experience you'd get with a subscription - with Art XML metadata and stuff (artist, date, etc). We don't always remember what a piece is or who painted it when it comes up..

Anyway, AFAIK, the only way to get TVs that aren't enshittified spyware is a business SKU, right?

In my house, we use Steam to play "windows-only" games on:
- Devuan with XFCE
- Devuan with Cinnamon
- Arch with hyprland
- bone stock Ubuntu 24
- ubuntu 25 laptop w/ second GPU

From my POV, there's not much need to port games to Linux. With the heroic efforts of Valve, most Windows games now just work. Win32, DX, D3D, and whatever else windows game devs have been using seems to have become the defacto reference gaming API on Linux.

Steam makes it work on every linux distro we've tried.

In writing this, it occurs to me: The F/OSS ecosystem does a very good job of re-implementing someone else's API/products (WINE, Proton, LibreOffice, etc)

The F/OSS ecosystem does a comparatively poor job at independently developing its own technology and then standardizing/universalizing those choices. E.g. the transition from X11 to Wayland; the systemd "situation(s)", desktop environments... gui greeters, audio muxers...

I think Valve has done the right thing. They made existing games work on Steam; they made Steam work on most linux distros.

Making everyone use a reference linux platform seems to be a total non-starter.

We already have a reference gaming platform: Windows 7 thru 10. And what we learned in 2025 is that Steam on nearly _any_ Linux often implements that windows reference gaming platform better than Windows 11 does.

I'd be more worried about zero day workweeks brought about by AI.