House House explains Big Walk's limited LLM use

A selection of IGF jurors said they would not vote for games using generative AI in any capacity; House House has now disclosed that an LLM was used on a small slice of Big Walk's code.

House House says it used LLMs sparingly during the final stretch of development to work through technical challenges and bugs. The studio originally drafted its disclosure as a suggested part of Big Walk's submission for the next IGF Awards, then published it after the use became a matter of public interest.

The studio estimates that well under 0.1% of the game's code had anything to do with LLM use, and says generative AI played no other role in development.

House House's example is a shadow-caster issue that made Big Walk run 50% slower for around 20 seconds at sunrise and sunset. After a Unity engineer concluded the desired culling behavior was not possible, an LLM chatbot suggested a possible fix. The developers extensively modified it and used it successfully; the fix amounted to about 10 lines of code inside URP. That is the most concrete part of the disclosure: a specific performance problem, followed by a fix the developers revised themselves.

House House says Steam does not request disclosure of code or chatbot use, so it had not disclosed those uses on the platform. The studio stresses that its disclosure is not an endorsement of AI tools. It supports efforts to limit generative AI's harms to videogames and the world, and apologized that Big Walk's use might not meet those efforts' standards.

House House called the lack of an internal AI-use policy a failure and says it will work toward a more comprehensive policy. The studio also worries that voluntary disclosures could make generative AI seem normal or inevitable, a concern that sits uneasily beside the jurors' stated refusal to vote for games using it.