he’s been salty about this for years now and frustrated at companies throwing training and compute scaling at LLMs hoping for another emergent breakthrough like GPT-3. i believe he’s the one that really tried to push the Llama models toward multimodality

i guess the point that’s being missed is that when i say “hard” i mean practically impossible

my point is that it’s hard to program someone’s subjective, if written in whatever form of legalese, point of view into a detection system, especially when those same detection systems can be used to great effect to train systems to bypass them. any such detection system would likely be an “AI” in the same way the ones they ban are and would be similarly prone to mistakes and to reflecting the values of the company (read: Jack Dorsey) rather than enforcing any objective ethical boundary.

but what are the criteria? just because you think you have a handle on it doesn’t mean everyone else does or even shares your conclusion. and there’s no metric here i can measure, to for example block it from my platform.

what about the neural networks that power the DSP modules in all modern cell phones cameras? does a neural network filter that generates a 3D mesh or rather imposes a 3D projection, eg putting dog ears on yourself or Memojis, count? what if i record a real video and have Gemini/Veo/whatever edit the white balance? i don’t think it’s as cut and dry as most people think

it’s already the case that the distinction between what’s “AI” and what isn’t is a subjective, aesthetic difference and not a technical one

pretty sure it’s SteamOS, an Arch Linux derivative, on a fairly popular Snapdragon platform. probably not too difficult to hack on it.

not really. using WASM as your full stack for your front end is just adding to the complexity and jank. WASM is there for compute heavy stuff. you can use it that way if you want.

i know i’m in the minority here so i’m not going to bury myself in this hole, but i do think those are addressable problems. many of them have been addressed. replacing Javascript is exactly what i’m talking about.

there may be a little angst from reading and rereading the “Max-Age” portion of the cookie RFC that caused this trauma

this is my most controversial take in computing in general:

i’ve always hated the browser. the reason there are only a few working browser engines is that HTTP and the HTML/CSS/JS tech stack is a gigantic pile of tech debt, and even using Chromium and Firefox you run into edge cases where, for certain edge cases, they don’t always follow the specs as defined in these ancient RFCs. and these specs: why tf are they treated as gospel? which software product specs drafted 50 years ago get this kind of reverence? why is it that other GUIs have had tons of iteration, not just of their spec but their full stack implementation (Wayland, .NET, Kotlin Compose, SwiftUI, etc), but we’re all just fine with this mess of janky boomer protocols cuz it lets startups get to market faster? why is downloading an entire app (less some caching) every time you want to use it feel less cumbersome than installing something native to the runtime environment where the protocols can be tightly controlled by the developer and not subject to whatever security and storage protocols whatever browser implementation decides is good for you? cookies? really? the browser should be reimagined with a tighter set of protocols that allow you to look at brochure sites and download content, ie apps. even the best web apps are a janky mess and have never worked better than properly developed desktop GUI. /rant

people generally probably hate the iOS integration just because it’s another AI product, but they’re fundamentally different. the problem with Recall isn’t the AI, it’s the trove of extra data that gets collected that you normally wouldn’t save to disk whereas the iOS features are only accessing existing data that you give it access to.

from my perspective this is a pretty good use case for “AI” and about as good as you can do privacy wise, if their claims pan out. most features use existing data that is user controlled and local models, and it’s pretty explicit about when it’s reaching out to the cloud.

this data is already accessible by services on your phone or exists in iCloud. if you don’t trust that infrastructure already then of course you don’t want this feature. you know how you can search for pictures of people in Photos? that’s the terrifying cLoUD Ai looking through your pictures and classifying them. this feature actually moves a lot of that semantic search on device, which is inherently more private.

of course it does make access to that data easier, so if someone could unlock your device they could potentially get access to sensitive data with simple prompts like “nudes plz”, but you should have layers of security on more sensitive stuff like bank or social accounts that would keep Siri from reading it. likely Siri won’t be able to get access to app data unless it’s specified via their API.