cs.LGNov 15, 2026

Weights & Caches: LLMs Can Program Tabular Foundation Models

Authors: Shi Bin Hoo, Magnus Buehler, Omar Swelam, Frank Hutter, Lennart Purucker

Organizations: University of Freiburg · Prior Labs · ELLIS Institute Tübingen

Abstract

Neural networks are typically programmed through optimization, rather than by explicitly specifying their computations. Recent work on weight-space programming suggests that large language models (LLMs) can instead reason about and manipulate neural-network weights directly. In this work, we show that LLMs can directly program statistical computations in tabular foundation models (TFMs), whose parameters enable inference across different datasets. We investigate three complementary settings using GPT-6 Astra, without gradient-based optimization. First, Astra directly programs nanoTabPFN from zero-initialized weights, achieving a mean ROC AUC of 0.849 across 20 TabArena datasets, compared to 0.876 for a pretrained baseline. Second, Astra diagnoses and repairs a signed-product regression failure in pretrained TabPFN-3.5, restoring near-perfect predictions on held-out synthetic datasets while preserving performance on TabArena-Lite, with only a 0.08% average relative increase in error. Third, it writes KV-cache-editing programs that steer TabPFN-3.5's predictions on unseen synthetic datasets, demonstrating transformations such as reversing linear trends and converting sine waves into approximate square waves, without modifying model weights. Together, our findings suggest that LLM-driven programming could complement gradient-based optimization across TFMs' development and usage, while offering a new way to investigate and intervene in their internal computations.

Figures & tables

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 17, 2026

Weight-Space Programming Doesn't Need Transformers

Standard neural architectures such as the Transformer were chosen because they train well with gradient descent and run fast on GPUs. In weight-space programming, a large language model (LLM) writes target weights by hand without gradient-based training, so those reasons no longer apply. Should we keep the Transformer? Given minimal guidance, ten labeled examples, and a sandbox, mainly on multiple-choice questions, we task three frontier LLMs with either generating the weights of a fixed architecture or designing the architecture themselves. When free to design, all three compile plain-text facts into a sparse table queried by a fixed procedure, reaching up to 54.4% test accuracy. An LLM-designed model achieves higher test accuracy than the most accurate released Transformer that the same LLM wrote while using about 1,800x fewer parameters. Without gradients, they choose a table, in which the LLM can state what each weight means and can find and rewrite a wrong one. Each wrong answer points to one row, and removing a row changes about one other answer. From this, we propose an associative-memory-based model up to 17x faster on GPUs than a same-sized Transformer. We also test pretrained weights, other benchmarks, and tasks where models generate answers. Pretrained LLMs remain more accurate; the claim concerns which structure to build when writing weights.
Nov 16, 2026

Replicating and Extending Weight-Space Programming: Frontier LLMs Compile Programs into Weights and Read Input Layers

Researchers at MIT recently reported that GPT-6 Astra, a frontier large language model (LLM), can write the weights of small neural networks directly in a single reply, and can read the raw weights of small networks to say what they were trained to do. In a three-day sprint, we replicated and extended these results. Our four findings: (1) Writing weights reproduces. Astra's networks for the game Breakout scored 514, 550 and 307 (versus MIT's 517), and scored just as well on two modified versions of Breakout. Claude Opus 5.5 can also write working networks, but often runs out of output tokens. (2) What is new is translating programs into weights. Earlier models could write a good policy as code a generation before they could write network weights, and the networks the models write behave like compiled programs: two models independently decompiled Astra's Breakout network into Python matching its actions in all tested states. On classifying handwritten digits, Astra's networks scored 50% on average, below a simple classifier (82%). (3) Reading weights is mostly reading the input layer. Astra and Opus 5.5 identify which game a network plays almost perfectly, but on three new games we designed and trained ourselves, Astra recovered where things are on the board, while mostly missing how the game works. (4) Auditing from weights alone is possible but shallow. Without executing code, Astra picked out a small language model poisoned during training in 10 of 10 runs, one poisoned by extra training afterwards in 7 of 10, and a malicious adapter in 23 of 32 cases, but never identified the trigger or malicious behaviour. Open-weight models lag behind. From this early study, we conclude that LLMs can read and write small networks, but that this currently amounts to compiling hand-designed programs and reading the layout of the input layer. It is a promising but early tool for interpretability.
Nov 5, 2026

Can an LLM Actually Reason and Program about Neural Network's Weights or is it just memorization on steroids? Exploring "Weight-Space Programming"

Large language models (LLMs) have recently been reported to read and write the raw weights of neural networks directly, a capability termed weight-space programming. We ask whether the reading side of this capability reflects reasoning about weight flows or a combination of local arithmetic at the network's edges and memorized priors over public datasets. We test Claude Opus 5.5 on small convolutional networks of about 125,000 parameters trained on five image datasets, providing every weight as plain text with no tools, no forward passes, and no dataset or class names. The model performs accurate local calculations: batch-normalization statistics lead it to infer about three optimizer steps on networks stopped after three updates, and output-layer sign patterns recover the vehicle/animal split of CIFAR-10. The sessions report reading the first and last layers closely and sampling the middle, and several claims built on partial tracing fail verification, although we did not run a direct forward-pass test. Semantic judgments lean on familiar datasets: the model names CIFAR-10 for CIFAR-10 networks but also for all three fast-food networks, relabels KMNIST as MNIST, reads a 99.57%-accurate SAT-6 network as CIFAR-10 animal photos, and calls "frog" the output favored by a feature that fires most on trees. Permuting CIFAR-10's labels preserves the output split at the new indices, but every class pair the model interprets is then wrong, indicating reliance on remembered label order. These are exploratory results from mostly single runs. We conclude that zero-shot weight reading by this model is better described as local arithmetic plus dataset priors than as weight-flow tracing, and we provide a protocol for auditing future weight-space claims.