Web developer. ☦️ Building @cruftjs Resolving representations since 1998
PNW
Joined July 2009
- Tweets13.8K
- Following792
- Followers2.1K
- Likes9.6K
Pinned Tweet
A 10 minute crash course on my theory of Generative Computing Construction Styles that you can use to build quite literally anything that is computable with an LLM. The Turing Machine, REST, and the Transformer can be derived from the same formal method.
All foundational systems software will need to be refactored. It’s all full of holes.
Everyone knows I started going by {First Name} in college. I haven’t gone by [First Name] since 2009!
An LLM is good at coding for the same reason you can get a sphere to go inside out without creasing or pinching it.
You can turn a sphere inside out without tearing, poking holes, or creating a single crease - and this is not a trick, it is a proven mathematical theorem. You cannot do the same thing to a circle. The reason is a topological invariant called the turning number, and it is the same invariant governing the geometry every LLM embedding space is built on. Stephen Smale proved eversion was possible in 1957. It took seven more years before anyone found a way to actually do it.
The mechanism is corrugation. Divide the sphere into guide strips connected by wavy sections that let the surface pass through itself without pinching. Push the caps past each other before a crease forms. Twist them in opposite directions - the same way a belt with a loop untwists when you rotate each end half a turn. Then push the equator through and uncorrugate. Bowls, domes and saddles cancel in a pattern that stays invariant throughout - the same structure that makes high-dimensional transformer geometry behave nothing like 3D intuition.
21 minutes. Bookmark & watch today. The mathematical proof that breaks every intuition you have about surfaces.
I've been thinking about how to think about building for the internet now that frameworks don't matter. REST is the architecture of the Web, but it doesn't say anything about how you should understand the design space of the client and server!
Software engineers en masse are just now discovering that you can derive software by decompiling it with AI. What they don’t know is that you don’t need a bajillion tokens and thinking set to ultracode. You can and have been able to do this with low thinking and frontier models from about a year ago.
I identified this capability in March of this year. I call it the Derivation Inversion.
Identify the invariants that induce the desired properties. State them in prose. Derive implementations from the prose. Every contingent choice the derivation makes is acknowledged as contingent; every property the implementation exhibits is traceable to a named constraint.
jaredfoy.com/resolve/doc/247…
You are decompiling Photoshop and clean rooming it in pure rust. I am disassembling pure rust and clean rooming it in assembly. We are not the same.
My JS engine’s JIT compiler has grown to over 160k LoC in the last few weeks, and most of that was a single rust file! Two days later it’s almost done refactoring!
Just feel the tension and unease in this room, only broken by the awkward applause of a handful of devs cheering on dhh’s wild-eyed enthrall as he delivers a eulogy to the computing priest class. 😱 I’m torn internally. I love what it used to be, but now I’m also shipping a self hosting rust compiler next week. 😵
It's right here. Properly edited. youtube.com/watch?v=vDjW_dRy…
Building software has never been more about the real humans using it than it is now. DX as it was understood is basically dead. Everything is now Agent Manager Experience; because Agent Managers are all being pushed to create value for others who are willing to pay for what Agent Managers do. The developer that wins this game is EXTROVERTED and CARES about their users more than anything else.
