Roman Pshichenko
6.3K posts

Roman Pshichenko
@Pshichenko
working on chat at SoFi. deep in ai. three kids.
Roseville, CA Bergabung Ocak 2014
98 Mengikuti1.4K Pengikut

It’s pretty straightforward. If you had all the resources and built computer use your way, then you would probably prefer to have complete control over the whole computer instead of just trying to guess what was happening on the computer by using screen pixels. Apple and Windows have complete control of the computer if they want it. If they build it into their system, then that’s how it’ll be and it’ll be much nicer.
English

computer use is broadly here & it’s genuinely very cool, but it’s worth flagging the structural asymmetry here where apple & google don’t have to pipe everything through accessibility apis unlike app players.
vertical integration lets them operate deeper in the stack like the compositor, view hierarchy, & event loop itself which is a real latency & reliability moat for on device agents.
the frontier labs are doing impressive work with pixels + a11y, but it’s a brute force path (almost hacky). that said, computer use today is still nontrivially difficult to actually use, & slow as hell (although combatted by background usage). the future is here, again just not evenly distributed to normal ppl yet.
ai using computers this way is an intermediate stage, definitely not the end stage. also before someone comes at me for saying this way allows long tail usage, it simply doesn’t matter because everything is power law where a handful of apps dominate time spent.
English

@tigerjwang @signulll I haven’t had time to try Kodak’s computer use yet, but it looks like it gets his own cursor and my understanding is that I can run apps in the background. I’m not sure how that works but I have seen some videos with two cursors.
English

my gripe with computer use right now: it's completely serial. GUIs/DEs were never built for concurrent operation because there's always one human in front of the machine. But agents don't have that constraint.
right now computer use hogs your entire screen. you can't even use your own computer while the agent works, let alone having multiple agents work concurrently on different tasks. unlike CLIs in which you can have multiple terminal tabs.
thought experiment: what if mac desktop spaces were independent agent workspaces - agents working in parallel across different desktops? that's probably the kind of OS primitive change that only apple can ship though...
English

@eagleed @mundoxrbrasil I wonder why it doesn’t look any different as you change resolution
English

First update for the YouTube app on visionOS (v1.01) is out. You can now scale the video window much bigger and video resume is smoother, which already improves daily use.
But it’s still missing key things… no native environment integration or lighting effects, so it’s not taking full advantage of Vision Pro yet.

English

@Austen Same until I realized pool maintenance, landscaping, even house cleaning are all services you can shop for
English

I used to see houses like this and think it looked so amazing.
Now when I see a house like this all I can think about is the maintenance.
Heshie Brody@heshie
honestly a TypeScript job in Tulsa Oklahoma might just be the move ($190–230K)
English

GPT-5.4 Pro solves Erdős Problem #1196!
Very pleased with this result; definitely my favourite thus far! This problem has been thought about for some time which makes this reasonably impressive and meaningful (see Lichtman's comments below).
Formalisation is underway!


English


@Pshichenko I ran the experiment. In my city none were available, but I found a set of plates about 150km away (just the plates, not the bar) for $1000 😩
English

@Pshichenko Here, this would cost as much as the "home" itself
English

@rtwlz If you’re a contractor and made $5M, then you were taxed $38 for every $100 you made. Compare to $25 or $16 if you made $400k or $100k



English


@ben_mathes @rtwlz If you’re buying 2.3 years every year then you’re doing something right
English
















