# Compiling Known Rules into Compact Neural Agents: Weight-Space Programming on Unseen Games

DevFeed: [Compiling Known Rules into Compact Neural Agents: Weight-Space Programming on Unseen Games](<https://devfeed.tech/articles/compiling-known-rules-into-compact-neural-agents-weight-space-programming-on-unseen-games-132298.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16981>)

Author: Jens Lehmann

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [games](<https://devfeed.tech/tags/games.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [neural-network](<https://devfeed.tech/tags/neural-network.md>), [ppo](<https://devfeed.tech/tags/ppo.md>), [programming](<https://devfeed.tech/tags/programming.md>), [python](<https://devfeed.tech/tags/python.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [weight-space-programming](<https://devfeed.tech/tags/weight-space-programming.md>)

## AI overview

In 20 privately authored game families, language models given simulator source could compile known mechanics into neural weights that generalized to held-out levels, renderings, and start states. With a fixed 57k-parameter recurrent network, the strongest writer reached 56% success with feedback; its Python players, and those of three other main writers, reached 97-100%. An exploratory free-architecture setup reached 68-90%, while support-example learners transferred less well despite receiving less information. The weights were sparse and program-like, but NumPy players under comparable compute caps scored 91% versus 79% for the networks. Training the architectures from scratch achieved about 1% within the tested budget. Noise-robustness prompts improved tolerance for two of three writers, yet PPO fine-tuning removed most success. The findings show rule-to-network compilation in this setting, with limited advantage over Python and no successful combination with tested PPO.

## Source excerpt

arXiv:2611.16981v1 Announce Type: new Abstract: Weight-space programming (WSP) uses language models to construct neural network weights without training. We test whether known game mechanics can be compiled into compact agents that generalise to held-out levels, renderings and start states within a game family. Writers receive simulator source with held-out level data removed for 20 privately authored, unpublished ARC-AGI-3-style game families. With a fixed 57k-parameter recurrent architecture, the strongest writer reaches 56% held-out-level success with feedback, while the four main writers' Python players reach 97-100%. An exploratory free-architecture condition reaches 68-90%. Learning from support-level examples transfers much less well, but these learners lack the source information given to writers. Written networks have a program-like structure: 93% of parameters are zero, and 90% of the rest are small integers or powers of two. NumPy players under comparable computation caps reach 91%, versus the networks' 79%. Training the strongest writer's architectures from scratch yields about 1% success within our budget. Written fixed-network weights are fragile, and standard PPO removes most of their success. Prompts for noise robustness increase tolerance by about four orders of magnitude for two of three writers, but PPO makes larger updates and still removes most success. These results demonstrate rule-to-network compilation in this setting, while showing limited benefits over Python programs and no successful combination with the tested PPO fine-tuning.