Aaaaaallan Deutsch

Aaaaaallan Deutsch

@allandeutsch.com

👨‍💻 Software craftsman 🎟️ SeattleJS Organizer Currently building: 🐤 docduck.dev - inline feedback for developer docs 🪟 paneform.com/wm - window manager for MacOS 🦺 typesafe-ts to help you write safer code with confidence

502 followers560 following2.1K posts

Aaaaaallan Deutsch reposted

✨ My new library gpuix-svelte is out! Build native desktop apps for macOS, Windows and Linux using Svelte and gpui, a Rust-powered graphics engine. Excited to see what you build! #svelte #native gpuix-svelte.stanislav.garden

gpuix-svelte.stanislav.gardengpuix-svelte: Build native desktop apps with SvelteA Svelte renderer for native desktop apps on macOS, Windows and Linux. GPUI, the UI framework behind the Zed editor, draws them, with no DOM or webview.

"Is X a person" is irrelevant; the downstream effects of it are not. Corporate personhood means companies have some rights, ie free speech (citizens united) and property ownership, but not the right to vote. IMO the interesting question is what rights of personhood does a class of "person" get.

Siobhán 🪏@shibbi.me

I cannot express to y’all just how many times I’ve been tempted to give up and say, yes, okay, fine, a thermostat is a person, let’s go

If LLM agents get personhood, what does that mean? - Do they get the right to vote? - Can they be sued for their actions or is someone else (operator/provider) responsible? - Can they own property? - Make political campaign contributions? - run for office? - right to due process? Which subset?

I think it’s always reasonable to be worried the other team’s candidate will win. Sometimes it happens. I mean shit we ended up with Trump twice! Given that, I’d prefer they run someone who I think is a competent candidate. Opinions differ on who is competent, but it’s better to care than not IMO

Henry @henrythedog.bsky.social

any ways if Kamala Harris is an unpopular loser flop who can't win elections why would you care if she runs in the Democratic Primary? like I don't care if Rahm Emanuel runs because he'll lose and it'll be funny. so maybe? worried she'll win?

This may come as a surprise, but FDA testing doesn’t require any understanding of how a drug works. The FDA tests that the drug has the claimed benefit(s) and does not have common and dangerous or severe side effects. So this thought experiment is how modern drug discovery already works.

SE Gyges@segyges.bsky.social

i hate this

Great to see so many new TLDs are on the horizon!

Pete Millspaugh@petemillspaugh.com

Today is ICANN Reveal Day! 1,615 strings: .agent, .api, .agi, .ask, .auth, .bad, .bro, ... 481 applicants: Google, Amazon, Meta, Microsoft, OpenAI, Anthropic, Cloudflare, Vercel, Squarespace, Airbnb, ... Gonna thread observations until I run out of steam

An old friend reached out today. He stumbled across one of my conference talks and wants to add it to the curriculum for a course he teaches. He recently started teaching at Yale, which made this even cooler 😎

I’m excited about this because if it’s easy to host it means we might have a nice path to atmosphere apps onboarding users without needing to route them through Bsky for PDS hosting. Every app having an easy path to onboarding net new users to the atmosphere will be a huge unlock!

Jaz@jaz.sh

Hi protocol friends! I've been cooking up a Very Large PDS (vlpds) implementation capable of scaling up to billions of accounts and down to just one. It uses Object Storage as its only state, supports HA operation, live scaling up and down, is feature-complete with the reference PDS, and is FOSS.

Aaaaaallan Deutsch reposted

Speaking to climate scientists this week, it was interesting to hear that we're likely to hit max emissions within the decade as renewable adoption picks up. If we sequester enough CO2 then we could make it back to 1.5 degrees. Fuck yeah! Let's fucking do it!

Kiran@kirancodes.me

what happened to like optimistic future of humanity-vibes leftism? feels like a lot of the discourse nowardays is almost conservative in how it seeks to curtail technological progress etc. I want to make humanity a better place for everyone. United together!

Shockingly good news about talking to LLMs

Matt Hodges@matthodges.bsky.social

There's growing experimental evidence¹˒²˒³˒⁴ that LLMs are unusually good at talking people out of entrenched false beliefs and conspiracy theories. ¹ pubmed.ncbi.nlm.nih.gov/39264999/ ² pubmed.ncbi.nlm.nih.gov/41179846/ ³ sciety.org/articles/act... ⁴ connecticut.adl.org/resources/re...

It’s trivial to generate the text of every 12 digit number addition problem. Odds it worth training on the data? 🤷‍♂️ probably not. To list all the problems tho: def print_all_sums(): for a in range(10**11, 10**12): for b in range(10**11, 10**12): print(f"{a} + {b} = {a + b}")

Colin@colin-fraser.net

what's more likely, that people at OpenAI secretly invented a way to compress plaintext by a factor of 43 trillion, or that a computer program can do addition

It’s been several years since I went around Stanley park, so this was long overdue! When I was a kid, I used to do it on a bike. Cool to be able to do it on foot now. I normally feel spoiled that Seattle’s waterfront is my everyday run, but wow! Vancouver’s waterfront is gorgeous! #running #kimbia

I could see rejecting a candidate for bringing alcoholic, but a coffee? What an inconsequential detail to focus on for such a big decision.

Kai Ryssdal@kairyssdal.bsky.social

I’m sorry, but if you bring an iced coffee - or any damn beverage at all - to a job interview, you’re disqualified. It’s not hard, people.

The difference in response to AI between software devs and artists fascinates me. Art values a work's provenance, whereas software is usually valued by its usefulness. Artists see AI as lost commissions, vs devs are happy to send people with "million dollar" app ideas to build it with chatGPT.

Singularity's Bounty e/CC@catblanketflower.yuwakisa.com

You have to consider very many people really haven't created much in their lives, and to make something as colorful and involved as this is a wonder Of course the haters hate it. But there is joy everywhere for those with eyes to see

I'm really curious what screenshots look like to LLMs. We disagree often enough about what alignment and gaps look like that I'm starting to wonder if they're seeing the same thing I am.

Okay tried prime agent for a few days, and it seems to consume dramatically more tokens than open code while producing worse outputs. Maybe I'm just using it wrong, but I thought it was supposed to use less tokens and do a better job; I've seen neither benefit.

Aaaaaallan Deutsch@allandeutsch.com

Taking prime agent for a spin after all the praise I've seen from @timkellogg.me and others. First impressions are pretty good! I feel it provides better insight to how subagents are being used, and like that subagents don't block main agent interactivity. Too early to judge quality/token usage.

Using built-in terminal colors (ANSI) on servers will send the color codes to my machine, so if theming is set on my machine remote TUIs render right & respond to my system light/dark changes. How did I not learn this sooner?! 🙃

I’ve been feeling the care for details degrading quickly the past 5 years, both from the transition of career bureaucrats to admin yes-men, and additionally the mass cognitive offloading to AI. I wonder how much the former has influenced the latter.

"Online Rent-a-Sage" Bret Devereaux@bretdevereaux.bsky.social

And the 'don't like book learnin' crowd is going to want to dismiss that, but of course a modern society runs on having engineers, accountants, doctors, scientists, *state departments* who treat the dry matter of their expertise with meticulous seriousness. It demands that approach!

Aaaaaallan Deutsch reposted

When I started working on wm, I quickly learned that the OS is why window managers feel wrong on mac. Mac is not a cooperative OS; it has opinions about where windows can move! To address this, I designed wm to learn how the OS behaves and adapt to it. Here's a comparison using learned Mac rules.

Aaaaaallan Deutsch@allandeutsch.com

I think the approach I’m taking is somewhat unconventional, I’m cookin’ another post on that which I’ll share later today.

When I started working on wm, I quickly learned that the OS is why window managers feel wrong on mac. Mac is not a cooperative OS; it has opinions about where windows can move! To address this, I designed wm to learn how the OS behaves and adapt to it. Here's a comparison using learned Mac rules.

Aaaaaallan Deutsch@allandeutsch.com

I think the approach I’m taking is somewhat unconventional, I’m cookin’ another post on that which I’ll share later today.

View full thread

Once found, the OS limits are saved. When wm observes the OS changing rule enforcement, it finds the new limits and changes to use those instead. By reducing its exposure to the OS, wm has a smaller surface for OS updates to break it.

When I started working on wm, I quickly learned that the OS is why window managers feel wrong on mac. Mac is not a cooperative OS; it has opinions about where windows can move! To address this, I designed wm to learn how the OS behaves and adapt to it. Here's a comparison using learned Mac rules.

Aaaaaallan Deutsch@allandeutsch.com

I think the approach I’m taking is somewhat unconventional, I’m cookin’ another post on that which I’ll share later today.

These aren't entirely unreasonable decisions from MacOS, since it provides no way to recover a fully off-screen window via the GUI. Window managers want to "hide" windows that aren't part of the active workspace. Mac only allows this via spaces, which have a slow animation and lack quality APIs.

Mac window managers work around this by parking windows _almost_ off screen - the bottom corners leave the least amount on screen when a window shouldn't be visible. Hard coding the OS rules means buggy behavior when the OS updates its opinions. Instead, wm dynamically finds the OS limits.

Taking prime agent for a spin after all the praise I've seen from @timkellogg.me and others. First impressions are pretty good! I feel it provides better insight to how subagents are being used, and like that subagents don't block main agent interactivity. Too early to judge quality/token usage.

After trying many options and not finding a window manager I love on Mac, I decided to build my own wm. It seemed a waste not to share it though, so I decided to make a website for it. The website needs to explain the product, which is hard. IMO show > tell, so wm runs in a browser too! 😅

Aaaaaallan Deutsch@allandeutsch.com

I'm working on the coolest website I've ever made and I'm really excited to launch it soon.

This is an important read and I’m thrilled @seldo.com has a plausible approach for tackling the sustainability that has been ravaging open source.

Laurie Voss@seldo.com

I have been writing this post about the economics of open source, on and off, for 13 years. I wasn't expecting to end up at "Microsoft should send corporations a huge bill" but nothing else we've tried has worked. https://seldo.com/posts/nobody-pays-for-open-source-we-can-force-them-to/

Yeah if the unsubscribe link isn’t in the email or requires more than 1-2 clicks when I get to the unsubscribe page, it’s a “report spam” from me instead. 🤷‍♂️

Annie Sexton@anniesexton.com

if you make me login to unsubscribe from your email list I will report you to the police

Aaaaaallan Deutsch reposted

Replying to @pfrazee.com

Most devs agree all the other devs are bad at coding, so if an LLM trained on code from all the other devs surely it is also bad at coding.

In an era overrun with "mandatory service fee", "booking fee", "regulatory response fee" and others, one man dared to be different. I've run past the exquisite creatures exhibit a bunch, really excited to finally be checking it out!

Been trying to figure out why I find the AI discourse so off putting. It reminds me of sex ed as a kid: - anti-AI is preaching abstinence, no exceptions burn in hell - pro-AI discourse is shaming virgin and discussing the details MIA: the AI version of “here’s some condoms, have fun & be safe”