The primary use case for House House’s use of LLMs was to research, via chatbots, issues we encountered with third-party code. If you’re technically-minded, and curious, here’s an example. We needed to cull shadow casters by distance from the camera, but Unity URP only allows culling shadow casters by distance in camera forwards, so shadow casters behind the player were being rendered at an infinite distance. This made our game run 50% slower for around 20 seconds at sunrise and sunset. We researched the problem extensively, and we looked at the URP code ourselves, but we couldn’t find a way. Then, we were able to speak directly to an engineer at Unity, who asked around and ultimately concluded that what we wanted wasn’t possible. Finally, we posed the question to an LLM chatbot. After a few dead ends, it suggested a possible fix, and while we had to modify the suggestion extensively, it worked. This fix represented about 10 lines of code inside URP.

"Temptation" is one way to put it, I suppose, when the human experts tell you it's impossible. This award's criteria was already decided, and they're entitled to their definition, but this is just information that was surfaced to them rather than work done for them.

I'm not sure this is a great policy. You can tell the difference between using ai for debugging and a whole game made of ai slop. Nobody in that second category will be just sneaking by unnoticed.

But using ai in the final stretch to squash some bugs, that's pretty hard to notice from the outside. If you outlaw the use of ai entirely, what do you get? People still use it for the little things, but don't admit to it. Is that better?