kevmo314

At that rate, they might as well have gone one step further and made the x axis exponential scale to make it feel even bigger.

You iterate on what’s shipped. It’s not a one-and-done kind of deal.

I don’t get it. I clicked the demo link but there’s nothing to try, only a home page with some AI generated images?

Even if this were legal it feels like a pitch for a scam…

This sort of thing is part of the problem. If it takes reading such a long manual to understand how to properly use Git, it's no wonder everyone's workflow is different.

They don't have to be, nor am I claiming that all nighters are unilaterally positive. But they were an integral part of my college experience and many of my friends' and I enjoyed them in a type 2 fun kind of way.

Asserting that they're not worth it misses the broader picture.

The course I did best at in school was the one that led to a job opportunity through some friends I made.

> All-nighters are not worth it.

I disagree. I made some of my best friends through all nighters and continue to occasionally pull them because they reinvigorate meaning into my work as they did my coursework.

If your only metric for success in school is your GPA, then yes all nighters aren’t worth it. But climbing a metric leaderboard isn’t the only measure of doing well in a course.

It is curious because Andrej recognizes this with his comments concerning coffee.

This is a very clever insight, nice work!

Since there’s monitors that work over USB-C, could one replace the capture card and somehow get the laptop to pull video like a USB-C monitor?

What is teamwork if not a feedback loop?

I would be pretty impressed by anyone who managed to do that nominally. Moving a dozen billion dollars alone seems not trivial to do.

Given that four of the five are Asian languages, there's a lot of transferability. Not a crazy amount but enough to give you a boost. Knowing Chinese made learning Japanese feel a notch easier, and learning Korean afterwards felt yet another step easier.

The 2200 hours represents learning from scratch.

This makes it difficult to configure Caddy in anything except the native Caddyfile language due to a lack of thorough documentation. It's an interesting idea, but configuring Caddy with a yaml config that someone prior deemed a great idea was quite painful.

Curiously, LLMs have made it a lot easier. One step away from an English adapter that routes through an LLM to generate the config.

Then people will keep building tools like this. Which, there's nothing specifically wrong with, but rather it's clear there's a demand for ffmpeg-but-more-ergonomic and ffmpeg is in the best position to do that.

With the recent Whisper integration there's already precedence for a way to do that.

I wonder if it would be more productive to bake that into ffmpeg at some point.

For example, something like

> ffmpeg -llm "can you take input.mp4 and resize it to 720p, transcode to vp8 output to output.webm"

Am I just a package manager?

I was able to get a little bit faster than the multithreaded version with a single thread using page-aligned reads and Grand Central Dispatch for asynchronous read operations: https://github.com/healeycodes/counting-words-at-simd-speed/...

This was a nice opportunity to learn about Grand Central Dispatch through AI for me as well. I knew about the page alignment and async read techniques but not on OSX.

Edit: actually you can get even faster with mmap https://github.com/healeycodes/counting-words-at-simd-speed/...

The ecosystem really is the main argument. If you want to hire developers who can work productively in your codebase, part of that is being able to match their expectations of what the code looks like. React standardizes your codebase, even if the standard produces complex code.

Yes, I believe that's basically what https://serpapi.com/ is doing.

Yeah, I was about to say, I sure hope they do.

I've been working on a better tab completion model that stays as an extension: https://ninetyfive.gg/

The main feature I really care about is low latency, which is my main gripe about Copilot. There's still a ways to go to match Cursor's quality but I'm chipping away at it!

This is more of a usability problem. In the second example, it's obvious that `json.dumps()` blocks everything else and it can be readily observed. It's not obvious that it blocks in the former and I've encountered many surprised coworkers despite it seeming obvious to me.

I think a lot of people assume you can slap `async` onto the function signature and it will not block anything anymore. I've had PRs come through that literally added `async` to a completely synchronous function with that misunderstanding.

Isn't this "echo with more steps"? The CI/CD example [1] strikes me as not obviously better than doing

          cat > .env << EOF
          DATABASE_URL=${{ secrets.TEST_DATABASE_URL }}
          STRIPE_API_KEY=${{ secrets.STRIPE_TEST_KEY }}
          EOF

which also addresses the trust and rotation problems. I suppose for dev secrets those are annoying, but even with secretspec you would have to rotate dev secrets when someone is offboarded.

[1] https://devenv.sh/blog/2025/07/21/announcing-secretspec-decl...

But why though? That post seems to leave it at

> After five years, there can even be tax benefits to this kind of accounting.

without actually explaining what those benefits are. I'm genuinely curious why one would choose to amortize.

Interesting, that's about what I suspected. Makes sense despite the absurdity, thanks!

> Companies have the choice to amortize salaries if they want.

I am curious, is there ever a time you would want this? Maybe if you’re operating at a loss?

You've missed the point of the article, which in fact agrees with your anecdote.

> It's equally common for developers to work in environments where little value is placed on understanding systems, but a lot of value is placed on quickly delivering changes that mostly work. In this context, I think that AI tools have more of an advantage. They can ingest the unfamiliar codebase faster than any human can, and can often generate changes that will essentially work.

Anything that involves sharing data with other people will run into issues around updating. If your API surface is a shipped sqlite db instead of an API call it's liable to be abused in so many ways.

What's the incentive to finish the marathon? 0.05% of a 1% chance of a billion dollars?