akurilin

I've heard that kind of description used for Switzerland as well. Accurate?

Superwhisper trains their own S1 models for this, you can use either a smaller local version of them or a beefier one in the cloud

Really liked that "Go Full Screen" as a modal flow, surprisingly intuitive.

Curious which model this can supplant as a clear winner on almost every metric. Sol? Looks like it's not quite there on a couple of benches, but I'm not clear how much they matter in practice.

Couple of related reads/watches to this I found useful recently, with a similar conclusion:

https://www.youtube.com/watch?v=FUUzmRH5Yi4

https://www.construction-physics.com/p/robot-dexterity-still...

It feels like systems like the Maturita' final multi-day exam in Italy do a pretty good job at providing students with a standardized, state-supervised process for demonstrating one's competence by the end of high school. No computers, you're under the watchful eye of your teachers the whole time, nobody knows what's going to be on the exam until the day of.

https://en.wikipedia.org/wiki/Matura#Italy

It's time consuming and nerve-wracking, but it feels fair.

"few applicants, most pretty good" was true for us as well when hiring Haskell developers at Freckle (https://tech.freckle.com/) - in some ways it was scary: we're away from the herd, the numbers aren't on our side, what if we ever need to hire large volumes of people? Instead of getting 200 JavaScript devs in the pipeline, you would get maybe 10-20 Haskellers. What if the person you're looking for is not in that pool? Intuitively, you assume that more is always better.

In other ways it made filtering through candidates quite simple: half of the people who applied were really good, the other either language astronauts or folks fascinated with the tooling who didn't genuinely want to move fast to produce commercially-viable software, they wanted to tinker. You just needed to figure out which bucket the person was in.

Cue https://worksinprogress.co/issue/the-housing-theory-of-every...

pg wrote about this a while back: https://www.paulgraham.com/determination.html

Came here to see how it performs against Luna. Would love to see how they compare on specific types of tasks.

I enjoyed this magazine and subscribed to it for a while, but I wish there was a US-centered version of it for those of us on this side of the pond.

Really hoping that all of the attention you're bringing to the longitudinal sloppification of codebases makes it back to the labs and creates some pressure to improve that trait of the models. This new benchmark seems promising.

At the same time, I imagine it will be hard for them to prioritize this over improving flashy one-shots of impressive zero-to-one feats that demo so well and attract more customers.

Can we have simonw make "pelican on a bicycle after 1000 requests for iteration" popular?

Greg Puciato's Fuck Content album was expressing a similar sentiment: https://gregpuciato.bandcamp.com/album/fuck-content

Do you happen to have any links that document this? I haven't been able to confirm.

I took a swing at bringing this into Handy.computer if anybody's interested: https://github.com/cjpais/Handy/discussions/1031 . Looks like there has been past demand for someone to implement it, but no proposed PRs. This article was inspiring.

Would this end up replacing the default iOS keyboard dictation functionality in iOS 27?

Reminiscent of the Vibecession analysis done by Scott Alexander a few months ago: https://open.substack.com/pub/astralcodexten/p/vibecession-m... - may be good supplemental reading

And of course the evergreen Housing Theory of Everything https://worksinprogress.co/issue/the-housing-theory-of-every...

Good read. I would love to see a breakdown of what exactly these businesses do, see if there are any trends here.

Oh, and thanks for writing Game Programming Patterns. That was a great resource.

Very relatable. I still recall when our team at Microsoft, about 20 years ago, had a dedicated "build engineer" who would be given the task to create a new branch for the product. He would disappear for a few days and re-emerge with the announcement that a new branch had been officially created and was ready to be used. Hard to imagine that in git world.

When I went back to Perforce I was reminded of why that was a thing. You're making a full copy of everything. They recently introduced a virtual copy stream, but I remember people advising against using it, for reasons I can no longer remember. It seemed like a potentially sensible approach.

Couldn't agree more. I wonder why the company stalled as much as it did. To be fair they've been around for three decades, maybe it's just hard to keep momentum for that long, especially if you're in a dominant position with no competition coming after you.

Thank you! It was a bit of culture shock for me when I first experienced it, coming from the world of business software, and I figured others would find it equally fascinating to learn more about.

We had to use Perforce (Helix Core Cloud) at my last game studio, and it is the de facto industry standard that most of your creative staff is already familiar with. The programmers don't love it, but they don't rule the roost in games. It's also the safe, verified default for working with Unreal Engine 5.

It does show its years though. We were one of the first users of the Perforce cloud offering, as we were small and didn't want to self-host ourselves, but it was a bit of a rickety experience. You had to register an Azure account in order to be able to access the service, and you had to ask support to modify things like triggers. Coming from the world of GitHub and other SaaS products, you could tell it was an attempt to retrofit an older model into a new skin.

The Git LFS path has some unofficial support as well, but you are on your own when things go poorly. Epic doesn't provide much help there.

Competition in this space is welcome, especially if they're planning to make it fully officially supported by the Engine.

I wrote about why merging files isn't as common in the world of game dev for folks coming from the world of text: https://www.kuril.in/blog/why-game-devs-dont-merge-files/

Looking forward to reading this. Still have a signed copy of The Lean Startup from an event in Seattle from 15 years ago. The book had a big part in pushing me towards doing my own lean startup just a year later, so, thank you.

I own two Glove80s, but for me the ultimate split keyboard with a thumb cluster was the Dygma Defy, of which I now own a pair as well. Couldn't recommend it enough. https://dygma.com/pages/defy

The designers were extra mindful of different thumb lengths, making it so that you can usually find a configuration that feels natural and not cramp up your joints: https://dygma.com/blogs/product-development/how-we-developed...

Curious what those two hours would entail!

> Another reason is that on the supply side, nobody wants to sign up to do a bunch of free work just to be rejected. If you just put up work, the candidates incur all the risk, meaning they walk away with nothing if you don’t hire them.

It's true, but prepping for a typical senior+ onsite loop in big tech still requires weeks of grinding leetcode, re-learning the latest system interview questions and the system interview answer framework, refreshing and rehearsing STAR stories, studying the company and its unique quirks that you're expected to know to pass the culture filter, remembering how to do all of this speedrun-style since you only get 40ish minutes per session, etc.

While that knowledge is more reusable across onsites, it's likely even more work than doing real or pretend-work for the company for a couple of days.

> When candidates get to walk away with something of lasting value that they can keep forever

I'm curious why them getting rejected from the position, even with the work sample they can carry away with them, wouldn't be still interpreted as a negative from future employers. "The other co passed on them, am I the fool for thinking they're good?" type of herd mentality which is often unavoidable.

Won't that "work sample guest book" be treated as the list of all companies that rejected you, a net negative for your personal brand you're projecting?

> (Me paraphrasing what Steve was implying) Take-homes are impacted by AI one-shotting them for candidates

I've been pleasantly surprised by how much you can glean from having the candidate upload their conversation log with the coding agent for whatever take-home you give them.

Author here. A collection of personal experiences of working with business co-founders and CEOs as an early stage CTO in the Bay Area VC-backed world. Hope it's useful to others in the role who want to maximize their chance of success in this critical partnership.

This is great stuff, walking the reader through your thought process was helpful for me as a developer to grok why yolobox was designed this way. I ended up landing in the "just make a local copy, don't get fancy" world myself after many iterations of workflows. Separate agents, separate containers, separate ports, that all resonates.

You mention this approach gobbling up a bunch of extra disk space as a consequence of the tradeoffs. Have you considered using APFS cloning on macOS to reduce some of that burden, or is that too tiny of an optimization to be worth it at this point?

Great link, thanks for sharing. Confirmed what I saw empirically by comparing the different models during daily use.