Ants in the Machine
Communicating swarms of small LLM agents will become as omnipresent in our digital environment as ants are in our biological one.
Posts tagged: ai
Running LLMs in the Browser with Three.js
A tour of how language models generate text, how GPT-2 and newer Phi and Qwen architectures differ, and how Three-LLM implements their inference graphs with Three.js TSL compute shaders.
An M5 Max using Samsung's LPDDR5X-PIM would generate 650 tokens/second on an 8B model
Replace the M5 Max's LPDDR5X with eight Samsung LPDDR5X-PIM packages and the memory could deliver 4.9 TB/s where the weights live. The right parallel mapping could push Llama 3.1 8B toward 650 tokens per second, or a 27B dense model toward 190.
Processing in Memory: DRAM Is About to Do Math
Samsung's LPDDR5X-PIM puts compute inside commodity memory, exposing 614 GB/s of bank-level bandwidth for local LLM inference. The silicon works. Quantization, memory layout, and runtime support now stand in the way.
Neural Texture Compression in Three.js
A small multiresolution feature grid plus a tiny MLP decoder that replaces a stack of correlated PBR textures with one compact, GPU-decoded representation. Named channels with their own activation functions, built on new fp16/half-precision support in TSL, and demonstrated on a MeshPhysicalNodeMaterial. The format is material-agnostic.
Neural network basics, for graphics people
A short reference on the neural-network vocabulary used across the neural texture, neural material and neural appearance posts — MLP, weights and biases, ReLU, forward pass, loss, gradient descent, backpropagation, Adam, latent vectors, feature grids and decoder networks — explained once, concretely, so those posts can link here instead of re-deriving it.
Zstdify: Pure TypeScript Zstd
Today, I built a pure-TypeScript Zstandard implementation (library + CLI + dictionary training + interop tests) in just 4 hours using OpenAI Codex 5.3.
Origins of Exocortex
The story of how I coined the term "exocortex" in 1999 to describe a synthetic cognitive regulator: a technological layer that joins the brain's regulatory hierarchy and helps govern attention and reasoning.
From Agents to Intent
AI coding agents are great at greenfield tasks -- but struggle to maintain or modify existing systems. The problem isn’t the model. It’s the paradigm. This essay makes the case for intent-based programming: a shift from task-driven implementation to declarative specification, powered by systems like Declary.
Generic Builder
The Generic Builder is a configurable meta‑builder that creates specialized builders on demand, solving the chicken‑and‑egg problem of intent‑based toolkits. This essay explains its design, the router agent, and a chat‑driven workflow that bridges conversational intent with structured files.
Recursive Intent
Recursive generation allows intent-based programming toolkits to ingest their own outputs and build entire applications from minimal specifications. This essay explores how recursive intent turns code generation into a compiler-like pipeline that scales small prompts into structured systems.
Sufficient Specifications
Traditional specifications failed because they required exhaustive detail. With LLMs, we can now embrace "Sufficient Specification" - providing just enough intent to guide generation without drowning in details. Learn how iterative specification specificity and generator patterns make intent-based programming practical for the first time.