I've simplified it and made it more flexible

3._1_415926535897932384626433832795_0_288419716939

In terms of aeronautics, went from the Wright brothers to the moon in 40 years. After that, everyone understandably thought that we'd be living in space and flying everywhere with personal jet packs in another couple years. Little did they know, it was the top of the S-curve, not the middle.

In the 60 years since, we've barely been able to adapt the 737 to fly longer routes.

Sure, I have no doubt that AI progress will follow an S-curve. The question is, where are we on it and is the plateau at a level safe for humanity? That's a very difficult thing to estimate without a crystal ball and not a risk I want to take.

Lol I can't tell if this is sarcastic or not

At the same time, the only things that get built and sold are the things that someone is capable of building and selling

I believe that the issue was that the EU wanted Apple to open up their new AI agent interface (the ability to control every app on your phone so Siri can call you an Uber or whatever), and Apple thought that it was too risky of a capability to give to any random AI app right out of the gate.

> Apple thought that it was too risky of a capability to give to any random AI app right out of the gate

Oh come on. Apple doesn't want to give up control. That's what this is about. The privacy thing is just to make them look good

Nothing you wrote is in disagreement with the parent.

It's both but the dangers are far more issue for their brand than control

> There are 2 potential outcomes: either the sky really does fall, and there's a meaningful uptick in bad things happening to iPhone users, in which Apple can easily point the finger at the EC and say "they made us do this". Apple looks like the good guys who put up a good fight for their users, but ultimately their hands were tied, and they'll probably get the revisions to EU law they're so desperately fighting for.

I'd prefer they focus on safeguarding my data instead of playing a ridiculous game of brinksmanship with regulators to make a point.

I agree. Safeguarding data and user freedom are 100% compatible, no brinkmanship required.

The expectation confirmation theory[0].

Someone else’s unrealistic assessment frame expectations, especially when they are attempting to speak from a place of authority, which they were. When reality doesn’t meet or exceed those expectations it creates disappointment. The expectations they set were impossibly high.

This is a pretty common thing. I’m sure we’ve all been disappointed by a movie or restaurant that a friend hyped up endlessly, which really didn’t live up to the expectations that were set. It’s the same deal here.

[0] https://en.wikipedia.org/wiki/Expectation_confirmation_theor...

Pokemon Go was pitched as Pokemon on your phone with AR to integrate into day to day life. There's no reason to expect anything but the Pokemon games I've played, now natively on my phone with AR integration.

What came out was a clone of Ingress with a skin and a shop. It lacked the full set of Pokemon, which all the assets for already exist. It lacked having a six-Pokemon team. It lacked trading, a core feature of Pokemon in every generation of games. Gyms weren't even gyms, they were some sort of checkpoint XP farm thing.

If it had been pitched as what it was, I may have enjoyed it more. Instead, I found myself vastly disappointed with what I was able to achieve playing it compared to Pokemon on my Nintendo DS or some other handheld console.

I don't think this was a politics-based decision. I feel misled and disillusioned.

This is not uncommon when becoming disillusioned with something that has been hyped up and forced upon you for an extended period.

The fatigue of the product (and sting of false promises) causes the negatives to overshadow anything positive to say.

These guys may actually just be angling to sell off the training data. diverse training data is more valuable

Also, cleaning kitchens is a huge part of the job. Hotel rooms either have no kitchen or a very minimal one. You're not going to learn how to clean an oven or load a dishwasher in a hotel room. (And loading a dishwasher requires categorizing thousands of things as dishwasher safe or not! Stainless steel skillet, yes; cast iron skillet, no; etc.)

Yeah, this seems like a much more likely option. Get a ton of good, completely unique scans of real world environments you could never replicate in testing and even if your product sucks and you fail entirely, you’ve got a really good dataset to sell to a big company that’s close on a product and needs data to enhance/refine on.

does it even work?

surely not. surely these coding agent tools wouldn't wipe data without asking for permission. surely no developers would be so incompetent to allow them to do that. (the buck stops with those devs.)

I feel like a lot of people take the guardrails off entirely, especially so you can wander off and come back to a PR.

The horror is if you're not running that in some sort of sandbox.

From my back of the envelope analysis for my own projects, paying per token on OpenRouter is competitive if not cheaper than running the same open weight model on a rented GPU. Per-token pricing is in the same ballpark (although more expensive) for closed frontier models and open weight models (cents to dollars per million). To me this says that the pricing is somewhat grounded in reality.

Are you comparing single-user requests or multiple concurrent requests when you say comparable to rented GPU? Most of the cost efficiencies kick in with concurrent/batch requests. A single H100 node can provide like 5k input + 2k output tok/s on a model like Qwen 3.6 35B-A3B with 30+ concurrent requests.