Probably for tasks that are laborious to do, but easy to verify; not sure what percentage of such work is at hand here.

LLMs are not super well suited for this kind of task.

Genuine answer: sometimes. I've done similar levels of reverse engineering. Keep in mind that LLMs, and all other AI, "think" as they are programmed to. When you need to do something they don't know how to do, because you're the first person in history to do it, it turns into a lot of hand holding. Especially when you're dealing with physical devices that can explode if misprogrammed, I've found it's best to keep them working on the lower hanging fruit.

They've really strict policy on LLMs. They pretty much don't allow using them, because the slop so much in this kind of region.

Nothing is stopping you from using LLMs when contributing to their project (I think). One reason might simply be that they would rather spend the (very sparse) donation money on anything else but tokens.