


Founder and CEO of F1R3FLY.io. Inventor of rho-calculus. Mathematics is an art. (ze/zer)





But in that vision you really need a notion of branching time non-determinism.
People thought I was crazy when I wrote “Computing without Processors” in 2011. I’ve long dreamed of one way to program everything that computes. You’d write a single program and compile it automatically to any mix of hardware and software, with the split shifting to fit your performance and cost targets. That’s been hard. Hardware and software use different languages, semantic models, and abstractions. The people who build them come from different university departments with very different cultures. These divisions exist because of how humans organize themselves, and because of the limits of human brains. AI doesn’t care about any of this turf marking. It relentlessly synthesizes hardware and software together, integrates them, and formally verifies the result, or makes it correct by construction by deriving it from a formal spec. In some cases we may not need CPUs at all: an application’s control logic gets “partially evaluated” and dissolved into the computation itself. I’m hoping to get the last laugh, probably sometime in 2027. dl.acm.org/doi/10.1145/19785…

i can now state exactly what's been bothering my about both Lean verified software and transformers. It's exactly one phenomenon and it's the linear-time versus branching time divide made manifest. Of note: naturally evolved intelligence evolves in an environment full of races: which sperm fertilizes the egg; which antelope gets culled by the lion pride; which flowers gets pollinated by the bees. This is branching time behavior. Naturally evolved intelligence makes use of races. Lean and transformers can't. That's a mathematically provable statement. github.com/F1R3FLY-io/public…

Fractal tiling of 60°-90°-120°-90° kites, from my 2001 paper in Computers and Graphics, “Fractal tilings based on kite- and dart-shaped prototiles". This particular fractal tiling was previously used by Peter Raedschelders in an Eschereque drawing. More at mathartfun.com/encyclopedia/…

Today i signed an executive order declaring that dark matter is where all the aliens are. The Fermi Paradox is simply their cloaking tech. My cabinet has unanimously agreed that with a single stroke of the pen i have solved two of the most vexing problems in physics. Thank you for your attention to this matter.

Happy Sunday! Here is a version of rho with a notion of symmetry built in (github.com/F1R3FLY-io/public…). There's a notion of stabilizer. There are applications to fault tolerance and error correction. One key take away for programmers is how to program one level up. Think of rho not as a single calculus, but as a type constructor. Then think about parametrically modulating the type constructor.

F1R3FLY now has an installer for MacOS (Apple silicon only). It installs our new f1r3lang native browser as well as a shard and starts the shard. github.com/F1R3FLY-io/ign1t1…


For Friday night fun i showed how to use rho with graded where clauses to score music. The key idea is that a note only happens with a COMM event happens. Channels are enriched with a timbre and a noteDatum. A noteDatum is either a pitch or a duration. Channel and co-channel are pitch and duration or duration and pitch, respectively. Thus, for( _ <- <@0, piano, A3>)0 | <@0, piano, qtr>!( 0 ) ~ piano A3 qtr ~> 0 plays an A3 for a qtr note using a piano sound. The graded where clauses allow one to express distributions over pitches and durations, which give musical idioms. Similarly, graded where clauses allow on to express physical constraints of specific instruments. For example playing pianistic stacked thirds is very hard, physically, on fifths-tuned string instruments. Here is a paper describing the calculus (github.com/F1R3FLY-io/public…). Here is a spec for an implementation (github.com/F1R3FLY-io/public…). Here's a Rust implementation of a player that takes a score as a textual representation and plays it (github.com/F1R3FLY-io/F1R3Sc…). Here's a draft slide deck summarizing the work (github.com/F1R3FLY-io/F1R3Sc…). i am finding that because rho with graded where clauses is so small i can just ask Claude to implement it as a basis for whatever application i am interested in developing. Of course, it would be easier to have one place where everything is implemented. But, this is a cool new way of working. Keep your computational abstraction small and compact and close to the domain you're working on. i keep finding that rho with graded where clauses is suitable for lots of domains.



Finally! We have a compilation path from rho calculus to GPUs. It’s a “short exact sequence” so to speak. First we compile rho calculus terms to rho combinator terms augmented with a new operator. Then we get rid of the new operator. Then we compile the resulting flat term to a matrix representation. This representation can be executed in terms of matrix operations executed on the GPU. Here’s a summary of the state of the work. Links to source code repo and papers in the comments. drive.google.com/file/d/19l7…

Guess what? Translating a rho calculus term into the rho combinators doesn’t blow up the term the way it does translating lambda terms into SKI or translating pi terms into Yoshida’s combinators. Reflection helps keep the translation linear. Meanwhile we have a separation theorem providing a way to get fine-grained control on the expressiveness of terms. See more here: github.com/F1R3FLY-io/public…