What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

Weight-Space Programming: LLMs Reading and Writing Neural Network Weights

Frontier LLMs can directly read and interpret neural network weights to understand what networks do, and can write or edit weights to optimize network behavior without traditional training loops. This capability, termed weight-space programming, enables tasks like identifying bugs in model weights and improving performance on specific tasks with minimal weight modifications.

rikiyat.github.io →

Connections

Supports

WorldTUI: Training a 360M Model as a Terminal Emulator WorldTUI trains a small model for a specific terminal emulation function; weight-space programming could potentially write such specialized small models directly, bypassing the training process shown here
Coding Agent Continual Learning: Analysis of ARC-AGI-3 Problem Solving Coding agent continual learning via ARC-AGI-3 explores agents improving themselves; weight-space programming offers a mechanism for that self-improvement by directly editing weights rather than fine-tuning
Dust: Pretraining Transformers Without Backpropagation Both challenge the necessity of gradient-based training; Dust eliminates backpropagation while weight-space programming replaces training loops with LLM-directed weight edits, converging on non-gradient optimization paradigms

Related to

NESformer: NES CPU Compiled into Qwen3.5 Weights NESformer encodes a CPU's functional logic directly into transformer weights - an extreme case of weight-space semantics that weight-space programming would need to parse and reason about

Develops into

Language Models that Play Chess and Explain Their Moves LMs explaining chess moves represents behavioral interpretability; weight-space programming is a deeper evolution - interpreting and modifying the weight substrate itself rather than just observable outputs

Type:	link
First published:	Monday, 2 November 2026 at 20:01 UTC
Location:	https://rikiyat.github.io/blog/weight-space-programming/
URL:	https://mattwood.fyi/i/0cdb96fa

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections