weight-space-programming

A blazing-fast, memory-safe neural network library for Rust that brings the power of FANN to the modern world.

The interface between probabilistic model checking and data-driven policy learning.

wsp-kit by Aurem — an engineering tool for small models. Try to fix small open-weight models by editing weights directly, no retraining. An eval gate decides.

First release for Can LLMs Debug Robots by Editing Their Weights? Code, 65-row edit, 16 experiment logs and evidence. V2 and real-robot experiments in progress.

Weight-space programming experiments on nonconvex algorithmic tasks: code, recorded results, analysis, and research preprint.

Agent skill for weight-space programming: edit LLM weights

Can an LLM write a neural network that IS a benefits formula, exactly? Weight-space programming experiments on SNAP, EITC, SSI, and Medicaid arithmetic.

Controlled weight-space programming evaluation with CPU replay, fixed neural networks, and audited LLM pilots.

Weight-space programming toolkit for inspecting, editing, and evaluating open-weight language models.

Calmer Qwen: fixed sparse weight edits, reasoning-language analysis, reproducible tools, and positive and negative experimental evidence.

Agent skill for neural-network forensics, causal debugging, and reversible weight edits, inspired by Weight-Space Programming.

Direct NNUE weight edits without gradients: reproducible Stockfish matches, 264 game records, and an inconclusive first result.

A curated list of weight-space programming: LLMs that read, edit and write the weights of other neural networks directly. Updated daily.

Direct weight surgery for LLMs — edit weights, not gradients. Extract the arrow, leave the body intact.

Awesome Weight-Space Programming: curated papers, code, benchmarks, and open experiments for reading, writing, and editing neural network weights.

Awesome Weight-Space Programming: curated papers, code, benchmarks, and open experiments for reading, writing, and editing neural network weights.

Circuit Injection for Weight-Space Programming

Give AI agents tools to inspect, edit, and test model weights: an MCP harness and Python toolkit for weight-editing experiments.

Independent replication and extension of MIT CSAIL's "Sparks of Weight-Space Programming" on Qwen3.5-0.8B: rebuilt loop edit, truth-gated reflex in 16 MLP neurons, pre-registered tests of knowing vs. committing. Report v0.3 (EN/DE), code, data, pre-registration log.

Multi-Agent Residual In-Context Learning: LM-Guided Mechanism Inference from Tabular Data