Hacker News: pkulak

New comment by pkulak in "Bijou64: A variable-length integer encoding"

pkulak — Fri, 29 May 2026 16:23:45 +0000

UTF-8?

New comment by pkulak in "New York passes pied-a-terre tax"

pkulak — Thu, 28 May 2026 17:30:04 +0000

Yup, very "Georgist", which I'm a huge fan of. You can move your money to another country, or hide it entirely in stocks that you borrow against until you die. But, you gotta live somewhere. Land is the only thing the state really has, and it's limited; it's the best thing to tax.

New comment by pkulak in "Constraint Decay: The Fragility of LLM Agents in Back End Code Generation"

pkulak — Mon, 25 May 2026 05:46:37 +0000

If we were doing novel things, we’d be scientists. I’m an engineer though. I don’t think I’ve been writing slop for 30 years.

New comment by pkulak in "Defeating Git Rigour Fatigue with Jujutsu"

pkulak — Mon, 25 May 2026 03:13:53 +0000

Wow, that's really nice. I'm absolutely stealing that. Thanks!

New comment by pkulak in "Defeating Git Rigour Fatigue with Jujutsu"

pkulak — Sun, 24 May 2026 22:00:41 +0000

JJ is a whole different way to think about source control. The fact that you don’t need to run an agent just to use it is a nice bonus.

New comment by pkulak in "DeepSeek reasonix, DeepSeek native coding agent with high caching and low cost"

pkulak — Sun, 24 May 2026 16:23:53 +0000

Doesn't Pi Agent do exactly this? Assuming "append only" means they do some kind of compaction as well.

New comment by pkulak in "Codex is now in the ChatGPT mobile app"

pkulak — Fri, 15 May 2026 17:09:00 +0000

I just like that Openai let's you use your codex subscription with whatever harness you like. I prefer Pi, so that's what I use. GPT 5.5 xhigh feels equivalent to Opus to me, so there's no reason for me to be locked into the Claude Code cli. I use it off-and-on throughout my workday and never even come close to the pro limits.

New comment by pkulak in "We are retiring our bug bounty program"

pkulak — Fri, 15 May 2026 16:57:15 +0000

Personally, and this is just my wild guess; I think almost no one is _really_ going to learn to code again. That's not a slight on the next generation or anything; I wouldn't know how to code right now if LLMs existed 25 years ago. I love coding, but it's the _result_ that's the real accomplishment. The journey a bit... but having the result always just sitting there, ready to grab with no effort, is too much temptation to resist.

And as this goes on, folks who can run an LLM _and_ understand/criticize/rework/re-prompt are just going to get more and more scarce. Even using an LLM in my preferred style, where you guide the model through a long series of small steps, will fade away.

New comment by pkulak in "We are retiring our bug bounty program"

pkulak — Fri, 15 May 2026 16:42:32 +0000

Yup, this was my first thought. Tell an LLM that there's a bug, and it will _happily_ add 200 lines to the project, usually wrapped in if statements so that it all interleaves with existing code. Then it will write twice as many lines in tests, run it all, and be done. Your bug is fixed. All the tests run, and test coverage went up. Now do that a couple dozen more times. :shudder:

New comment by pkulak in "Lanzaboote – NixOS Secure Boot"

pkulak — Wed, 13 May 2026 15:05:55 +0000

> without signature or review

What are you on about now? I got _one_ of my projects accepted into NixPkgs a couple years ago and have never done it since due to the huge PITA it was to find someone with contributor rights to sign off on it. If I want to update it, same hassle. Now I prefer to just throw a flake in the root of the project and call it good, which actually works really well.

Wait until you find out that Arch has both secure boot and the AUR.

New comment by pkulak in "A.I. note takers are making lawyers nervous"

pkulak — Mon, 11 May 2026 18:01:58 +0000

I feel like if you trained better for "I don't know", it would drag down competence everywhere else somehow. Like, the strength of a model is exactly it's ability to grasp at straws and very often find the right one.

If you ask a good model something that makes no sense, it will tell you it makes no sense and it can't answer the question; so I know it's possible.

New comment by pkulak in "MPEG-2 Transport Stream Packaging for Media over QUIC Transport"

pkulak — Fri, 08 May 2026 05:37:58 +0000

They are still pretty widely use for HLS/DASH streams. You may not be dealing with them, but you're probably watching them.

New comment by pkulak in "Appearing productive in the workplace"

pkulak — Wed, 06 May 2026 19:53:52 +0000

But those giant models get the boilerplate correct the first try! You're totally right though. My favorite thing to do these days is to hand craft the code in the middle of the app, then tell AI to make me a rest endpoint and a test. I do the fun/important part. :D

Though, that's coming from someone who can't justify thousands on personal hardware and is instead paying $20/month to Openai. Might as well use the best.

New comment by pkulak in "Vibe coding and agentic engineering are getting closer than I'd like"

pkulak — Wed, 06 May 2026 19:20:00 +0000

I can't get used to vibe-coded projects on Github. One that I was using for a little while is about a year old, with 40,000 commits and 15,000 PRs. And it has "lite" in its name; it's supposed to be the simple alternative. There were so many bugs. I fixed one, submitted a PR, but it was off the first page in hours. It will never be merged. I moved to a different project with a bit less... velocity, and it has been way smoother.

New comment by pkulak in "Appearing productive in the workplace"

pkulak — Wed, 06 May 2026 18:18:58 +0000

One of the best uses of AI I've found is code reviewing stuff I've written either entirely myself, or even code generated in a previous session.

New comment by pkulak in "Claude Code refuses requests or charges extra if your commits mention "OpenClaw""

pkulak — Thu, 30 Apr 2026 15:49:30 +0000

I also assume that forcing usage to spread out, via those 5-hour windows, has cost advantages.

New comment by pkulak in "Claude.ai and API unavailable [fixed]"

pkulak — Thu, 30 Apr 2026 01:39:50 +0000

Not to mention substantially less open. I've been using an OpenAI subscription in Pi Agent for a couple weeks now and it's great. And from what I can tell, 5.5 is a heck of a model.

New comment by pkulak in "Waymo in Portland"

pkulak — Tue, 28 Apr 2026 19:01:56 +0000

Are Waymos cheaper than hiring a person?

New comment by pkulak in "Waymo in Portland"

pkulak — Tue, 28 Apr 2026 19:00:50 +0000

Same here, as someone who doesn't drive much, and is generally a "vulnerable road user". I've seen Waymos drive. When they screw up, it's by stopping dead under an abundance of caution. They never speed. They can spot a ped or cyclist from blocks away. Every time I take an Uber home, the driver is guaranteed to drive 40+ on the 20mph road in front of my house while blasting through crosswalks with people waiting to cross. The data is not really in yet (still not enough miles to really say if they are safer), but they pass the eye test.

The rain will be a real test though!

New comment by pkulak in "Waymo in Portland"

pkulak — Tue, 28 Apr 2026 18:56:56 +0000

Well, that's the hope, but the bar is pretty low. Portlanders constantly block streetcars, usually by doing a shite job of parallel parking.