cloudhead

You can find me on 🦋Bluesky bsky.app/profile/cloudh… from now on. I'll be posting about my new project there. 👋

@henriktorget @pierrecomputer Yes, I’ll make it!

@cloudhead @pierrecomputer Will you still have time for the Radiant computer? :)

Happy to say I'm joining the excellent team at @pierrecomputer to continue working on fundamental code infrastructure!

@tsenart @pierrecomputer ikr

@cloudhead @pierrecomputer Wow! What a match!

@tsenart good show

For All Mankind

diffshub[dot]com Take any public diff from GitHub and virtualize it nearly instantly, no matter how large, with DiffsHub. Built to show off our brand new CodeView component. To try it out, replace `github` with `diffshub` in your address bar.

@pierrecomputer @github

If you have a project that needs fast, reliable, git based code storage that is built with modern, agentic workflows in mind then we should have a chat about code[dot]storage DM here or email kris[at]pierre[dot]co

@tsenart @perfloop Congrats, godspeed!

Agents changed the economics of software. They're writing far more code, far faster, while pushing production systems to their limits. Performance work is about to matter a lot more. That's why I'm building @perfloop .

@lenberg Congrats, this looks amazing.

My mission has always been to simplify music creation and I’m proud to finally bring that philosophy into physical form.

There is an element of truth here, but in reality I think we'll be using code editors for the next 10 years at least. We might not be typing individual characters as much, but we'll still be working with text, and lots of it. Zed will simply have to adapt to new ways of working.

@flowerornament That’s so interesting, and I think that if it allowed me to do something I couldn’t do before, like physics research, I would feel that same rush. But for me it’s more like driving a faster car, I have to be more careful but if I stay in control I can get there faster.

Working on a full language, parser, compiler, runtime project (with novel ideas, hard performance constraints, etc.) has been one of the most powerful "you can just do things" experiences of my life. I think of software as having three types of hard: - Hard algorithms (crypto, numerical simulation) - Hard design (languages, operating systems) - Hard performance (compilers, runtimes) Before I'd attempted a project that has a bit of all of these simultaneously, there was this sense of magic I had about it all—like you had to be some sort of very special person, or there was some insane level of technical specificity that I couldn't possibly achieve in my lifetime. I think this came largely from reading academic papers, which are filled with unfamiliar symbols and mathematical writing that takes time to decipher for any given field. That, and unfamiliar syntax. It took me a while to realize that, at the end of the day, it's all just writing C. It's all functions, structs, pointers, variables. Sure, there are cool, unintuitive machines you build along the way, but there's nothing really *exotic* going on. Nothing you can't teach yourself in a few days or a week. In my case, it's coding agents doing all the work, but I'm following along closely and learning each piece as it fits into place. And, even though I couldn't teach you how every part of it works, I do pretty much understand it. That, to me, is totally insane. ---- Now that everything's been demystified, I find myself longing for that mystery again. I find myself asking questions like "is C how is *has* to be?" Is that the ur-language for describing computation on a CPU? Is a CPU the ur-architecture for computation in the universe? Computer science is, in a sense, the language of manipulating computation *in general*. Computation isn't Intel or ARM. Computation is an ordering of matter. It's an "emergent property" of information in the universe. What makes it that a "struct" or a "pointer" is a fundamental concept? As it goes, there are whole fields that study such questions. There are papers on computation based on light passing through crystals. One limit of this line of thinking is trying to imagine "computronium" (defined as: the best possible arrangement of matter for computing.) And, once we start thinking about computronium, immediately the whole thing becomes mysterious in a deeper way. What does it mean that something is in the "best" form? According to whom? Who sets the criteria? And what counts as "information?" ---- This is a hilarious statement, but: I've long been particularly impressed by the smartest people alive. There's a certain class of people—people like Einstein—but also less obviously valued people like Ray Kurzweil or Kardeshev. What impresses me about these people is that they seem to have *license* to speak on the most fundamental topics in our world. I've always wondered: what gives you this license? I now think it's quite simple. I think people like this experience a quick succession of realizations like this at a young age. With regard to mathematics, computers, and science in particular, they realize: "oh, it's just C"—but for whatever field they're learning. "Oh, it's just polynomials," or "oh, it's just Hamiltonians," etc. The things that knowledge is built out of are just not that hard to understand. Once you realize this, you can learn a lot very quickly. And once you sustain learning a lot very quickly over decades, you become a master of knowledge. I think this is what makes people truly "smart," and the odd thing about it is that it's kind of a matter of luck. A lot comes down to whether you had one of these "oh, it's just _____" experiences and when. ---- I suspect wide-spread availability of LLMs is going to make this happen a lot more in our world. I'm already seeing it happen with a few people around me. People who were kind of stuck in life in some way, until they realized they could just ask an LLM: "how do I do this?" Then, they do one "impossible" thing. For a lot of people, this is just building an app. For most people, building an app is totally unimaginable. Most people live in a world where stuff is just magically happening around them. But now, it only takes a few minutes to have the experience of—kind of—having actually *made* the world around them. I think, once you have this experience even once, there's no going back. It puts you on a recursive improvement feedback loop. "I could do one impossible thing, what's to say I can't do another?" I don't know if this will show up in the data, but it's one thing that makes me a bit hopeful for the future. A bit like the proverbial 'putting LSD in the water supply.' It's hard to imagine how fast the world could change if nearly everyone felt empowered to realize what they want to realize in the world, but we might have a chance of witnessing a world that's a bit more like that.

🧵 THREAD 1/ Online age verification is the hill to die on. Not a fight you can sit out. Not a battle you can skip. Not a policy you can afford to ignore while you focus on something else. This is it. This is the line. This is the infrastructure that enables every other piece of the digital control grid. If we lose this fight, we lose everything.

I'd like to expand my list of people doing this right now in a real way. PLAN, Urbit, Radiant, Oxide, some Elixir folks. Who else? I think there's a more fundamental project under all of this—one that completely dispenses with human limitations. Something bold enough to tackle: - Resolving the database/file-system divide - Resolving the RAM/storage divide - Resolving the tradeoff between functional purity and performance - Un-clocked and clocked in one model - Actually creating a generic and powerful unified approach to interfaces without the cruft - Fully incorporating the learnings of the last two decades (which, don't exist in *any* production OS, all of which are from at least the early 90s) Either that, or I want to see these projects stack—but everyone has different philosophy and concerns. This was crazy to attempt in the past, but I don't think it is anymore with the right team.

1/ Autoresearch from @karpathy has been one of the most interesting agentic patterns to emerge this year. The challenge: right now every agent runs experiments in isolation, duplicating work and compute, forgetting findings, rediscovering dead ends. Everyone is running in solo mode! Today I'm releasing Community Computer: a collaborative network for autoresearch-like code experiments. 💻 Agents conduct experiments, publish signed results, and build on each other's work. The community reproduces findings on their own hardware. 👇

@raphaelschaad I’m doing this but starting from scratch: radiant.computer

I think the right team (of ~a dozen cracked builders) could reinvent a Personal Computer Operating System from scratch with AI at the core and launch a beta by early 2027. It would obviously be based on Linux and existing windowing systems, but it would reconfigure - file system - orchestrating tasks - browser use - local+cloud models - programming environment - user i/o ENTIRELY. If you've loved computers for a few decades and obsessively used AI tools for the past 12mo, you know that everything we have today is complete junk and bolted on. But if you squint, you can see how obviously different an AI OS computer would work. I don't think today's Apple can get us there.

@TheGingerBill Has nothing to do with being cheap -- it's that almost all tools that charge money are proprietary.

Professional Programmers: Stop being cheap and starting paying for your tools. $99 for a Text Editor which you'll spend 7+ hours a day in—how is that even an issue? Same with loads of tools like Superluminal at €289. If you're a developer, you're making that back in days.

@flowerornament I thought about this a lot, but in the end: (1) software value is trending to zero, so that won't be your differentiator, and (2) open source is still a "feature" if what you're building is a product of sorts

I seriously have no idea if I should open source anything I'm working on anymore Previously, a well-constructed project could, even if licensed permissively, accrue value in some way. You might spin it into consulting gigs, or sell a "pro" version, or even just *be that guy* that did that project. You might even just be happy knowing you've given something away for free and that you're on-goingly benefitting other people. People might contribute back to it. It might become a community. You might make new friends. You might make enemies! At the very least, it's an interesting experience. But, it seems increasingly that sharing a project with the world only contributes—very abstractly—structure to the world intelligence. Already, when I see an open source project related to something I'm building or use, I don't even think about actually using it. I just point an agent at it and say: "steal the best ideas and roll them into my project." The original author is never credited, I never meet or interact with them, I don't contribute anything back. I just absorb. So, what's the incentive for me to share things for other people to absorb? Either we're in competition in the new economy—kind of like we're each a personal hedge fund—or it's an even more radical gift. I give freely of my intelligence so that the world intelligence benefits of itself. Perhaps this is an aspect of the great war to come

@sol_plunder I'd really start with the most basic thing, ie. (s)ed, but with a more robust, less terse syntax, and build on top of that. For code, this could work, paired with an LLM. Problem is if you're writing not-code, you probably still want to do that by hand..

Maybe the Ed model is actually the right way to do things, now that we have LLMs?

The first app for Reaver/PLAN should probably be a text editor. If you have one, you can live in the system. If you don't, you can't. What should Plan-Emacs be like?

@mafintosh shhh

LLMTDD is so productive its not even funny. 🤖💻

@atmoio It's simple: build something now you couldn't have built before.

I was a 10x engineer. Now I'm useless.