2026
GenMLX: Vectorizing the Full Inference Stack
This morning’s post was a snapshot at 10,800 lines. Since then, another 23 commits went in — vectorizing the full inference stack, adding gradient estimators, neural network integration, and compiled Metal inference loops. The theme: make GenMLX fast for real workloads.
GenMLX at 10,800 Lines
GenMLX crossed 10,800 lines of ClojureScript today. A quick snapshot of what went in over the last session.
Toward a Lambda Calculus for GenMLX
The GenJAX POPL 2026 paper (Becker et al., doi:10.1145/3776729) introduced lambda_GEN — a simply-typed lambda calculus that gives the Generative Function Interface a formal foundation. It uses quasi-Borel spaces for its denotational semantics, defines simulate and assess as source-to-source program transformations, and proves vectorization correct via logical relations.
GenMLX: From Prototype to GenJAX Parity in a Day
Formalizing Psychoanalytic Theory as Bayesian Multi-Agent Reasoning
Tonight I’m writing about something different from the probabilistic programming posts earlier today. This project sits at the intersection of my two fields — psychology and computation — and it’s the one I’m most excited about.
GenMLX: Gen’s Generative Function Interface on Apple Silicon
I built GenMLX — a probabilistic programming language in ClojureScript implementing Gen’s Generative Function Interface, running on Apple Silicon GPUs via MLX.
GPU-Accelerated Probabilistic Programming in ClojureScript
ClojureScript in a Blog Post
Can we run ClojureScript directly inside a blog post? Let’s find out.