cs.CENov 13, 2026

Interpretable weight-space programming builds transparent transformers and brain encoding models

Authors: Yash Mehta, Chandan Singh, Weijia Xu, Richard J. Antonello, Adeen Flinker, Nima Mesgarani, Michael F. Bonner, Nebojsa Jojic, 

Organizations: Microsoft Research · Johns Hopkins University · Columbia University · New York University

Abstract

Neural networks are powerful predictors, yet their opacity limits their usefulness in science, where the goal is explanation rather than mere prediction. In neuroscience, for example, deep encoding models accurately predict brain activity in response to natural stimuli, but struggle to reveal which stimulus features drive those predictions. Here we introduce interpretable weight-space programming (iWSP), a method in which an LLM agent iteratively builds interpretable models by explicitly programming their weights and features in code, without gradient-based training (backpropagation). Without backpropagation to tune parameters, the agent must draw on its prior knowledge and reason over previous trials to hypothesize, hand-craft, and refine human-interpretable computations mirroring the workflow of a human scientist. LLM audits confirm the interpretability of the hand-crafted models. We first validate iWSP by tasking it with constructing transformers for canonical benchmarks such as arithmetic and sentiment classification, finding that LLM agents can successfully craft compact, accurate networks directly. Next, we apply iWSP to build neural encoding models across vision, language, and audition. In the visual cortex, iWSP outpredicts existing interpretable baselines, uncovers distinct computations across early/higher visual areas, and generalizes to unseen data. In language fMRI, iWSP surfaces a confound in the data and in auditory ECoG, it systematically discovers interpretable acoustic features that predict neural activity. Together, these results establish iWSP as a general paradigm for aligning high predictive performance with human-interpretable scientific discovery.

Figures & tables

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Jul 3, 2026

Individual Parameters in Weight-Sparse Transformers Appear Interpretable

A central goal of mechanistic interpretability is to understand how neural networks work and what each individual component does. Dominant circuit-finding approaches focus on a specific behavior and reverse-engineer the role of components on the associated sub-distribution. However, past work has shown that components can have different functions that are active on different subsets of the input distribution. In this work we ask whether a single weight can be understood globally across the full training distribution by characterizing when it matters (the inputs on which ablating it changes the model's predictions). We introduce an automated LLM pipeline that writes a short, human-readable description of when a weight matters and verifies it on held-out text, crediting a weight only if its description generalizes. Across two sparse and two dense transformers, the fraction of weights that are interpretable (in this sense) is higher in sparse transformers than in dense ones, a gap that widens once unreliable descriptions are discarded. Our results show that a meaningful fraction of a sparse transformer model's weights can be interpreted: 12 to 31% of weights have a single short description that identifies what the weight is used for.
Nov 13, 2026

WeightBench: Evaluating Language Models as Weight-Space Programmers

Recent work shows that large language models (LLMs) can operate directly on the parameters of other neural networks. They can read what a network was trained to do, fix a network that fails, and write working networks from scratch, all without gradient descent. This capability has been termed Weight-Space Programming (WSP). Exploring this new paradigm systematically requires a robust benchmark. We introduce WEIGHTBENCH, a benchmark that evaluates LLMs of increasing capability on 42 WSP tasks across five tracks: reading, fixing and writing weights, recognizing edits that are impossible, and data-free compression. The networks whose weights the LLMs read and edit range from 2K-parameter control policies to 7B-parameter language models. Experimental benchmark results show that frontier agents can already read small networks and write working control policies without ever running them, but cannot yet patch language models, compress them without data, or tell an impossible edit from a hard one. On all 42 tasks, Claude Opus 5.5 reaches a macro score of 0.535 (Sonnet 5.5 0.466, Haiku 5.5 0.402, DeepSeek v4-flash 0.215); on the 38 tasks it was run on, GPT-6.1-sol reaches 0.568 against Opus's 0.546. Code, the task registry, our OS-level sandbox harness, and the live leaderboard are publicly available at https://github.com/Aayush-Marasini/WeightBench.
Nov 10, 2026

Mixed Signals: Weight-Space Programming for EEG and ECG

Weight-space programming allows language models to read a trained neural network's weights and edit them. This opens the door to ``training'' by reasoning rather than through gradient descent. Prior work has demonstrated its functionality and promise for game-playing networks and fixing specific failures in language models, even extending to diffusion models. In this work, we present its applicability to models trained on biosignal data, specifically EEG and ECG data. We show that frontier models are not only able to debug specific faults planted in biosignal-based models without being told what these faults are, but are even capable of writing all the weights of a network from scratch, given only the architecture, the task, and a minimal set of training subjects. In particular, in one set of experiments, a language model was able to match standard gradient-descent training using 6×6\times fewer training subjects, through reasoning alone and without any gradient descent step. Networks written by Claude Opus 5.5 also scored higher than classic hand-crafted EEG pipelines (alpha band power, and common spatial patterns with linear discriminant analysis) when built from the same 5 subjects. We also highlight some of the limitations, which include the inability of language models to improve already well-trained classification models, and their incapacity to reliably recognize, from the weights alone, the task a network was trained on. Overall, our results indicate that weight-space programming can be useful for small networks, such as those deployed on edge devices and wearables, but the concept still requires further investigation before it can be used in practice.