There’s a world of possibilities between “public facing server” and “turn the internet off”. Like using a VPN or mTLS. Or only connecting when you are at home.

However it sounds like you aren’t worried about the increased threat landscape, or at least aren’t willing to compromise on your current level of convenience. That is a fair answer to OP’s question.

Defining a threat model means to qualify what sort of attacker you expect to face and what they are assumed to be capable of

Which is exactly what OP is talking about. 5 years ago if you were only trying to defend against script kiddies, using containers was good enough. But now you have to consider that script kiddies might find container escapes.

I’m explaining why AI (at least local AI) is indeed deterministic. Same inputs give you the same output

The compiler can use rng too. From the gcc docs:

Sometimes gcc will opt to use a randomized model to guess branch probabilities, when none are available from either profiling feedback (-fprofile-arcs) or __builtin_expect.

More info here: https://stackoverflow.com/questions/52974259/what-are-some-examples-of-non-determinism-in-the-c-compiler

I’m willing to bet many people are not even aware that modern compiler optimizations use prng. Not to mention incorrect usage of the term “determinism” when referring to CS concepts. This is exactly what Linus is talking about when he cited lack of understanding of compiler behavior. I do believe there are notable differences between AI and compiler, but a lot of the reasons people give against AI are reactionary and not well thought out.

You can at least avoid using the spyware apps by using Matrix bridges

Replaying the seed will only get you the same result using a particular engine and a particular random number generator

And the seed and rng are all part of the LLM. If you use the same inputs you get the same output. That’s the definition of determinism.

Determinism does not care whether or not the user/programmer understands the program or not. The word you’re probably looking for is “unpredictable” or “less predictable”.

For local AI image generation, often the input parameters are indeed recorded into the output image. You can then use the metadata of the image to get the parameters and re-generate the exact same image. This is the definition of determinism in computer science.

All I’m saying is that determinism has a specific meaning in computer science. In an LLM, the rng value is determined by the seed. Given the same inputs, an LLM always produces the same output. It does not matter whether the user understands the code or operation at all. It is purely dependent on the program itself.

A non-deterministic program, is one where the output is determined by factors outside the code itself. This usually requires non-deterministic programming language constructs. For example, if you convert a set to a list in python, the order of the elements is non-deterministic. Out of the control of the program. Same inputs, different output.

For local LLMs you can set the seed and generate the same output every time.

If I ask "why did the LLM output “for” as the XXXXth word of its utterance, can Linus say why?

If you have the model that was used, and the seed used for the generation, then yes you can trace the weights and figure out how everything was generated.

AI is deterministic. And compilers can be just as “nondeterministic” if you don’t know what compiler was used and what flags were enabled.

By integrated into the OS, I’m talking about the major corporate implementations, like Apple, Microsoft Windows, or Google Android. The advantage is that you don’t need to install an app or browser extension, and then give that app permissions. Everything is already set up. I’ve seen some websites like Amazon try to automatically set up passkeys for you, so it’s likely that a good portion of the population already has passkeys set up in these corporate passkey managers, before they have the opportunity to learn about alternatives.

Speaking of Keepass, you should check out these discussions:

https://github.com/keepassxreboot/keepassxc/issues/10407#issuecomment-1994673954

https://github.com/keepassxreboot/keepassxc/issues/10406

In short, keepass had a way of exporting passkeys unencrypted to disk, and some people did not like that, threatening things like:

This implementation is not spec compliant and has the potential to be blocked by relying parties.

The unfortunate piece is that your product choices can have both positive and negative impacts on the ecosystem as a whole. I’ve already heard rumblings that KeepassXC is likely to be featured in a few industry presentations that highlight security challenges with passkey providers, the need for functional and security certification, and the lack of identifying passkey provider attestation (which would allow RPs to block you, and something that I have previously rallied against but rethinking as of late because of these situations).

This is exactly what the “passkeys are a trap” article warned about. Passkey attestation allows websites to reject specific clients, just like how SafetyNet allows apps to block GrapheneOS users. In the future, banking websites or google.com could force usage of passkeys, and block usage of keepass. Are there enough keepass users to protest this? Certainly hasn’t worked well for GrapheneOS and SafetyNet.

this seems to be an EEE scheme of Authy specifically

Functionally, it’s locked down because it’s nearly impossible to migrate the TOTP codes to another app.

You don’t really see custom TOTP apps because users don’t want to have to install an app just to generate TOTP codes for a specific website. Google Authenticator could have done this, but I think Google decided to start moving towards passkeys instead of enshittifying Google Authenticator.

There is much more incentive to do this with passkeys, because you can lock down passkey sharing in ways that can’t be done with TOTP. For example, Netflix could block passkeys from any client that allows passkey copies and exports, like Keepass, in order to block account sharing.

yup, and Netflix will use passkey attestation to make sure you can’t share logins with other users

Why did these corps implement TOTP?

To reduce support load from people getting their accounts hacked and asking support to get them re-instated. I don’t doubt that these incentives also drive the push for passkeys, but passkeys are much more opaque and tightly integrated with the OS, and is thus more vulnerable to centralization.

why aren’t these corporations locking down TOTP to restrict them to their own validator apps

They are. Authy is the biggest offender I know.

Edit: some info about how Authy locks you in: https://www.bentasker.co.uk/posts/blog/general/switching-to-aegis-for-totp-codes.html. Note that some websites integrate directly with Authy and force you to use Authy to set up the TOTP. The only way out is to use a rooted phone to transfer the codes to another app like Aegis.

Attestation for passkeys (software-based, syncable credentials) is antithetical to their mass market goals, and would freeze out independent, non-corporate-backed credential managers. It’s an explicit non-goal of the designers of passkeys, who work together in a forum anyone can join.

The web wasn’t supposed to be controlled by corps either, and look where we are now.

If passkeys were to follow the same path as the web, here’s what it would look like. Corporate passkeys get the most users because of OS integration and convenience. Now that they have all the users they can commandeer the standards, and make them so complex and evolve so quickly that non-corporate implementations can’t keep up (think Firefox vs Chrome). Corporations convince websites to drop support for outdated passkeys, citing security issues. Non-corporate implementations are locked out. Monopoly is established.

And now that passkeys are centralized to corps, the government can exert power on them, mandating hardware attestation for things like hardware-backed id and age verification. They push for backdoors, so they can log into anybody’s account. Bye bye privacy.

deleted by creator

Cloudflare also forces you to enable javascript, so allowing them to fingerprint and de-anonymize you

wouldn’t even go that far, the examples he mentioned are pretty uncontroversial at this point

Each album takes an hour or two. Faster if I can find a page somewhere with the whole album listing, preferably including the names of everyone involved with the album.

For OP this process would take 100+ hours for their collection. That’s an insane amount of work for just organizing your music.

How much free time do you people have? At that point I’d just use AI

Honestly…sounds like something an AI agent could do pretty well. Then when you’re browsing the library and you find a mistake you can fix things as needed.

Weigh the environmental cost against how much you value your time. Using local LLMs is probably enough for this kind of job, with much smaller environmental cost.

Using lemmy as example, when i request posts from another server, my server will request only those posts. It won’t request everything. Do you see the difference?

That’s just how activitypub works because the alternative would be too costly. But you could design a federation system that replicates everything.

A precise description is:
Usenet is a distributed system where servers peer with each other to replicate articles (P2P at the server layer), while users interact in a client–server manner with their chosen news provider.

Even your perplexity output admits that it’s not fully P2P like bittorrent. And if we count “partial P2P”, then activitypub would also count since even selective replication, is some amount of replication, and tbus “partial P2P”