Natanael

The judge explicitly did not allow piracy here. Only legally acquired media can be used for training.

Edit: https://www.techdirt.com/2025/06/26/two-judges-same-district-opposite-conclusions-the-messy-reality-of-ai-training-copyright-cases/

This case didn’t cover the copyright status of outputs. The ruling so far is just about the process of training itself.

IMHO the generative ML companies should be required to build a process tracking the influence of distinct samples on the outputs, and inform users of potential licensing status

Division of liability / licensing responsibility should depend on who contributes what to the prompt / generation. The less it takes for the user to trigger the model to generate an output clearly derived from a protected work, the more liability lies on the model operator. If the user couldn’t have known, they shouldn’t be liable. If the user deliberately used jailbreaks, etc, the user is clearly liable.

But you get a weird edge case when users unknowingly copy prompts containing jailbreaks, though

https://infosec.pub/comment/16682120

The ruling explicitly does not allow pirating. It only lets you run ML training on legally acquired media.

They still haven’t ruled on copyright infringement from pirating the media used to train, and they haven’t ruled on copyright status of outputs (what it takes to be considered transformative).

This is judge Alsup, same guy who ruled in Oracle vs Google

Edit: https://www.techdirt.com/2025/06/26/two-judges-same-district-opposite-conclusions-the-messy-reality-of-ai-training-copyright-cases/

1: SteamOS don’t run unnecessary services in the background. (especially stuff like print services and other random shit)

1b: Even regular Linux, which does run a bunch of extra services, still generally has less overhead because it’s still being optimized for lighter weight systems, and it idles more efficiently too. Meanwhile Windows doesn’t have a good way to tell your printer driver and its corresponding services to shut up when you’re gaming.

2: Antivirus programs

3: Drivers, graphics system. This is both a plus and minus, but for performance mostly plus. More efficient driver model, less overhead again. Sometimes the performance comes from lacking features which doesn’t get executed fully, though. Sometimes it comes from translating to Vulkan, because DirectX has some more overhead (and in these specific cases you can get the same performance boost on Windows by switching to Vulkan).

I run a cryptography forum

Encryption doesn’t hide data sizes unless you take extra steps

It’s called traffic analysis

Timing of messages. They can’t tell what you send, but can tell when

Or they’re trying to figure out who’s trying to stay connected with who

Consider getting VoIP phone numbers from a jurisdiction that’s much less hostile, so you have another number available to use

Proton is only for running Win32 binaries

Telegram also don’t have E2E encryption on groups

Do you think a device with regulation circuits is more likely to be overloaded and start fires…?

The infinitely easier solution is to let the car charger know how much power is available to draw.

Besides the general security risk of they run trojaned clients, if they run it in the office they’re spending the company’s electricity

It’s called incident response

I remember this lol

Tldr neural network models are incredibly weird. My best guess is that the combination of common recurring structure with variations based on common rules (joke threads and all) helps the model derive some intuition about how to handle variations of things.

Also reminds me of an even earlier neutral network which got better at playing specific games after being trained on large amounts of text completely unrelated to the game, like encyclopedias or whatever.

After the heavy criticism they changed it from default on to (opt out) to default off (opt in).

In theory you could modify it’s database, but they did mention applying stricter security (but what good does that do when the frontdoor access remains via the prompts)

Access controls is the big difference. Apps with sensitive data can choose to hide stuff to a system wid search API. It can do so on an individual level, even. And even if it previously was accessible it can be drumroll recalled. Exposure happens when a search is made.

Microsoft Recall is all or nothing. Once it has been displayed Recall has it and you can’t selectively erase stuff. Exposure is immediate. It’s just purge the whole database, or leave it all in there. Apps can’t retroactively flag stuff.

… But leaving AI summaries on by default was very stupid by Apple

Tech recruiters are the worst, almost nobody actually understands technology so they just pick based on the fanciest education and whatnot because they don’t understand how to judge experience

Maybe we could use this stuff for umbrellas too?

Not in the winter, lol.