HN user

rweichler

157 karma

twitter.com/rweichler

github.com/rweichler

Posts4
Comments83
View on HN
Claude Fable 5 1 month ago

That being said I think the harnesses are only getting better. And maybe we will get multi-modal models that understand architecture eventually. But the growing-the-blob-of-text training method that's being used now appears to be getting diminishing returns

Claude Fable 5 1 month ago

I can attest to this, I had a very simple 20-line shader that I asked Claude to do a basic 90-degree rotation on it, and it just completely got it wrong. Frequently adds pointless abstractions / intermediate variables even when I tell it explicitly not to in the system prompt. I can go on and on, these things just don't understand architecture. And why would they? They were trained on text.

There is something remarkable about turning speech into code (don't need to hunch over a keyboard nearly as much these days, can just talk into a mic) and it's good for first drafts / exploring ideas. But it's obvious to anyone that's paying attention we're hitting the top of the S-curve. It's no wonder the IPOs are around the corner. I mean even Dario admitted he doesn't know how they're gonna substantially increase the context window size. That says a lot.

Discord Incident 3 months ago

Times like this make me miss the IRC days, I was just able to reproduce a bug in an semi-open-source project, and Discord went down right in the middle of me sending my findings. Now there's nothing I can do about it. I can only wait.

Ghostling 4 months ago

For some things there is nothing better than xxd. It is so simple.

Ghostling 4 months ago

Yeah xxd is the correct answer. If you don't want to install a dependency and have Lua installed, or if you're just feeling a little bit frisky, you can use my function which is Production Ready™.

    Xxd = function(name, input)
        if not name:find'^[_%a][_%w]*$' then error('bad name: '..tostring(name)) end
        local ans = {
            'const unsigned int '..name..'_len = '..(#input)..';',
            'const unsigned char '..name..'[] = {',
        }
        local t = {}
        for i=1,#input do
            table.insert(t, ('0x%02x,'):format(input:byte(i)))
            if #t == 16 then -- 16 columns per row. arbitrary, change this if you want
                table.insert(ans, table.concat(t))
                t = {}
            end
        end
        if #t ~= 0 then
            table.insert(ans, table.concat(t))
        end
        table.insert(ans, '};\n')
        return table.concat(ans, '\n')
    end
I am distributing it under the terms of the GNU GPL v3. So if you put this in your codebase I will sue you into releasing your entire source. Just kidding it's MIT licensed.

Honestly that's a terrible joke. Seriously it's MIT. Here I will put the full license in this comment to illustrate how serious I am:

Copyright 2026 rweichler

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Why Objective-C 5 months ago

This is why I went with GNUstep. All of the APIs are 1:1 with Apple's, including the runtime (which is very important imo). It has AppKit support as well; its AppKit implementation is kind of half-baked, but at least it exists. Also, a lot of the APIs are kind of old. But looking at how Swift evolved over the years, maybe that's not a bad thing.

My main complaint with GNUstep is the licensing. The runtime itself is MIT which is great, but its implementation of Foundation/AppKit is LGPL. ObjFW, including its runtime, is LGPL. At least with GNUstep one day I can create my own version of Foundation based on Cocotron or swift-corelibs-foundation or something, and not need to muck with rpaths + ship a bunch of .dll/.dylib/.so files with my app in order to comply with the license.

I read this comment, about to snap back with an anecdote how I as a 13 year old was able to learn Lua quite easily, and then I stopped myself because that wasn't productive, then pondered what antirez might think of this comment, and then I realized that antirez wrote it.

As a complete noob to the space, as an individual, who doesn't care about being "hireable" or whatever, which one has the highest ROI? OpenTofu?

Srs question, I keep reading everywhere from experienced people Wayland sucks. I need to start learning of these stacks, should I go with Wayland or should I go with Xorg?

If I didn't know any better I would learn the Wayland API. Just like how: if I didn't know any better I would learn Swift (instead of Objective-C). But thankfully I do know better and I know to stay far away from Swift [1]. Is it the same deal with Xorg/Wayland? It seems like noobs prefer Wayland but the experts prefer Xorg.

1. https://youtu.be/ovYbgbrQ-v8?t=1456

I guess it provides less freedom/flexibility on the dev side. You're forced to use LuaJIT, and you're forced to buy into OpenResty's coroutine paradigm where global variables act weird, and stuff like that. Also I bet it doesn't play super nice with LuaRocks.

Totally valid choice to make, but in my opinion OP is missing out. OpenResty is state of the art and has a ton of great libraries embedded in it. It's "batteries included" so to speak, and the batteries are well designed. Yichun Zhang is one of the GOATs, along with Mike Pall. And Roberto, obviously.

Yeah, seems like the power law is at play here. I made my test harness in 2020 so I didn't have a choice as Lima didn't exist back then. I should have waited a year. I'll certainly keep an eye on it

Interesting, thanks. Looks much better than Docker/Podman. But seems to suffer from the same incentive issue. I think I'll stick with my raw QEMU setup, Lima seems like QEMU + batteries, but I already built the batteries.

On the topic of ditching Docker, has anyone else created a custom test harness with QEMU? I feel like I'm the only person doing it this way. QEMU's target userbase is emulators in general, which is a much broader audience with way more development effort going into it, therefore I don't think it can ever go "out of fashion" or get hijacked by perverse corporate interests like Docker can. Podman seems to have the same vulnerability.

This is now my 5th time writing the GUI part of Ghostty from scratch: [...] once on macOS with SwiftUI, then on macOS with AppKit plus SwiftUI [...]

Ha, that's a nice way of wording that. I'd take it a step or two further. :)

When Cook became CEO, all of this was inevitable. I used to blame Jobs for not picking Forstall as his successor, but it recently dawned on me that it was never his choice to begin with. The board probably crowded him out again, just like the Sculley situation.

In a month Apple will have been on autopilot for longer than Jobs was at the company during the 1997-2011 heyday. Jobs became iCEO in September 1997. After 167 months passed, he left in August 2011. It has been 166 months since then.

Looks cool but this project is only 11 days old. If this was mature I might've used this in EQE. But now I have my own bespoke solution.

I'm surprised how many lines of code this is, I guess the tagged pointer stuff adds a lot of complexity

You gotta remember, this was over 10 years ago. The VC/tech game was a lot different back then. The "search engine for apps" pitch kinda made sense back in those days.

A "search engine for apps" sounds stupid nowadays because it's impossible on iOS. But back then, jailbreaking was quite popular. Many Quixey employees (and many people in silicon valley) had jailbroken iPhones. iOS 3 was pretty limited and limera1n worked on every iPhone. Piracy was still mainstream. Spotify sounded stupid back then because Grooveshark was so much better. It was inevitable that Apple would cave in and open up its platform any day. And if Apple didn't do that, Android would eat their lunch.

Then Steve Jobs died, and Tim Cook released the iPhone 4S, the first iPhone you couldn't jailbreak easily. The rest is history.

Mobile was a new landscape, and it was a tough problem to solve. Plus they were going up against Google so there was a sense of FOMO if you weren't in on it. The thinking was, maybe there's a 1% chance it pays off and Quixey beats Google.