menacingly

I don't understand how anyone is still primarily thinking about single-user scenarios in 2026

I don't know the particulars, but in general, silence around a massive tech company on warrants does not mean "they said no and the feds decided to leave them alone"

stuff like this working is why you get odd situations like "don't hallucinate" actually producing fewer hallucinations. it's to me one of the most interesting things about llms

they seemed to enjoy the internet access in russia

This is great!

sum the amount of "you simply <x>" in this thread, then account for the fact that we're talking about running afoul of a regulation if you don't understand it, and you end up with a hassle. I'm not weighing in on whether or not it's bad, I'm just saying what I said. If you aren't accounting for a significant portion of revenue to justify it, you're going to get blocked because you represent a liability.

I'm surprised I don't see it more. You can't impose a regulatory burden more troublesome than your traffic is worth

Decentralized on centralized hardware?

could something like steam or a linux distro use it to distribute software? seems crazy not to.

I think what people are saying is we know there is probably a silent line somewhere, and we have no way of knowing when we cross it. It feels like it has an implied "most of the people reading this can treat it as unlimited in practice"

I'm not sure selectively enforced shakedowns instill a lot of confidence either

unless I'm misunderstanding, the same data could be pulled from those services.

the message content wasn't leaked here

You're only allowed to believe LLMs are a kind of digital messiah or that they're complete hype garbage, when of course the answer is some point between.

There is obviously a lot of potential here, but there is also a lot of solution looking for a problem.

My current red flag is if an argument hinges on a trademark breathless frisson for the growth potential. Statements like "models are getting smarter every month" that hasn't been true for a year. If your excitement over AI is based not on what we can do today, but a presumed future expansion for which we have no evidence, that's silly.

But what they can do today is cool. We worked for a long time to get computers to understand natural language intent, and LLMs demonstrably solve this problem.

it’s similar asking the gemini-1.5 models about coding questions that involve auth

one of my questions about a login form also tripped a harassment flag

it will defer to the server a _lot_, if you just consider the capability they can fit on that phone

if the servers are so private, why is on-device such a win? here are some irrelevant distractions:

- the cpu arch of the servers

- mentioning that you have to trust vendors not to keep your data, then announcing a cloud architecture where you have to trust them not to keep your data

- pushing the verifiability of the phone image, when all we ever cared about was what they sent to servers

- only "relevant" data is sent, which over time is everything, and since they never give anyone fine-grained control over anything, the llm will quietly determine what's relevant

- the mention that the data is encrypted, which of course it isn't, since they couldn't inference. They mean in flight, which hopefully _everything_ is, so it's irrelevant

microsoft was always going to take SMBs. Data is what makes them useful, so Microsoft keeps their SMBs, Apple gets consumers, Google gets their slice of productivity and android, where their preachy models will let you know if you did a harassment

The privacy conversation was pretty shady, and honestly full of technical holes with pointless misleading distractions

you should have seen them when they had a soul

it feels like microsoft showing off apple software

To a technical person, the discussion around AI felt really creepy and disingenuous, full of irrelevant distractions

this is the video form of "bug fixes and performance improvements", and about as charismatic

the more reasonable assumption is that it uses a third party crawl service.

but still, it doesn't matter, it's acting on behalf of a user, and you aren't entitled to know what software your users run. One consequence of putting stuff on the public internet is that it's like public.

we've all been there, go get 'em

I like the idea. Looks like you've got some kinks, happens to the best of us. I look forward to playing with it when you get those ironed out.

Also, I don't care if you respect robots.txt.

it's because there is some sense among the physics-challenged that apple has to support it in hardware, as if we're a generation away from placing the burn-inducing, fusion-reactor-draining h100s in people's pocket it would take to run a model that impresses them as much as the tech demos of today

that's a good point. it could also be the same reason everyone else is struggling: it's very hard to find compelling utility for LLMs. We absolutely will, but it's difficult.

Right now everyone is setting incomprehensibly expensive amounts of compute on fire, for basically free, in the hopes utility emerges

if you frame "a creative" as people who select an emoji

That's certainly one of the available takes. Mine is that I'd need something other than some exuberant posts to think that's market essential at this point.

Either it becomes an essential part of workflows, in which case it becomes commoditized and we're worried about whether or not they'll someday support the mouse, or it's hype and it doesn't matter if/when they implement it

Apple isn't behind on AI any more than they were behind on cloud storage. It doesn't matter when, it doesn't matter if it isn't SOTA, I don't know how many different ways I can word that they have your data. Google has your data.

model-as-service companies can have all the features they want, their models can't do anything meaningful without the data Apple, MS, Google have.

Apple is at zero risk that someone else will have your notes, entire app library with data, mail, calendar, messages. They hold so much leverage. Even if you use some other (so far nonexistent) secret all-model, what are you going to access it with?

also if you're not trying to improve the security of your product by running random binaries from the internet. I'm concerned at the inability to separate the concepts of "what it does" and "what it says it does".

The idea that whether or not it needs scrutiny is impacted by your goals with the software is... creative