stellz

stellz

@piss.beauty

club rat, technology sister, dancefloor diva. 🌹Bushwick music: https://plyr.fm/u/piss.beauty

7.9K followers1K following32.4K posts

stellz reposted

this is exactly the face a wook makes when they interrupt your cigarette break to ruin your fucking day

This post is unavailable.

any oomfs rocking a phone-based ssh setup with a tiny external monitor when on the go i already have termius on my phone & the machines on tailnet, but it seems like it could finally be fab to do android desktop w/ a tiny monitor & keyboard. i'm talking fits in your purse size, ~11" for the monitor

I miss fighting games, I miss playing sf 3rd strike religiously on my breaks with other qa people at my first office job forever ago - making bioware's (rip) star wars slop mmo

guys we have good speech models already Dessa is not that dangerous. you can literally voice chat with the major providers already

pretty much everything man

Social media loves talking about how everything is getting worse. But what’s something that you’ve seen get _better_ in your lifetime?

Mercury 3 is good btw. diffusion finally produces something that's not annoying to perceive

tailscale is like. fine and all but I was basically just doing this slightly less securely with dynamic DNS for quite a few years. I'm not really getting why y'all suck dick for it so bad

I still believe this is essentially true

stellz@piss.beauty

12 GANs stacked at different parts of the stack. problem? do more GAN

oh your model's code isn't perfect?just ask some other dummies to adversarially review it constantly. then review those reviews with another model and optimize them. just add more systematic adversaries bro I promise

not enough adversaries but you're out of steps to put them at? turn some of your reviews into a review panel of 2-3 different models & synthesize their findings together do follow up reviews after any changes that address feedback from 1st round reviews review the whole branch pre-pr

I hope everyone's private trackers that run on a fork of the ancient what.cd Gazelle codebase have updated everything

I still believe this is essentially true

stellz@piss.beauty

12 GANs stacked at different parts of the stack. problem? do more GAN

oh your model's code isn't perfect?just ask some other dummies to adversarially review it constantly. then review those reviews with another model and optimize them. just add more systematic adversaries bro I promise

I'm already doing all this but I'm sure I can find a few more places to shove in more adversarial review

stellz reposted

welcome to the future. computers can think now. also there is 4

look I read the original blog about the ts 7.0 port and i just want to say they were wrong to do it that way sorry this isn't even me conducting my usual anti-go crusade really, it was simply inferior to how it could have been for a bad reason

VoidZero@voidzero.dev

Introducing tsrs: a Rust port of the TypeScript compiler 🦀 We are not forking TypeScript! ◆ In close collab with the TypeScript team ◆ Up to 8x faster than TS 7.0 & ~70% less memory ◆ Runs in WASM (works in your browser & workers) ◆ Soon part of Oxc npx tsrs blog.cloudflare.com/typescript-r...

"oh it was easier to directly port because of how go is structured" I do not care you are microsoft. burn some fuckin tokens and git gud

felt great to replace my janky old mostly handwritten react mess that live-rendered markdown loaded direct from the pds, using equally janky custom lexicons it's now just ssg that posts the standard.site.document record before regenerating the site and deploying to cloudflare

has anybody diy'd a system for mitigating task/goal drift over long periods my review loop has been great for making the code much better but it seems like it needs some element of evaluating how each step helps accomplish the goal

I suppose I could just add another review panel member every 2-3 commits that evaluates for "on-goalness" / task adherence instead of code quality & functionality or make it part of the branch review that happens pre main merge, but it might be way more annoying to course correct by then

nix oomfs don't particularly seem to have computers that work more often idk

Isaac Corbrey@isaaccorbrey.com

The benefits of Nix are basically complete confidence in your system working at all times. It's great. The downsides of Nix are you have to read and write Nix. It's not great.