Do you think the low-level nature of ASM contribute to how successful the LLMs are with it?
I think also that a large/smart/fast LLM can automate testing things with QEMU for at least moderately common architectures like motorola 68000, throw code at it rapidly and try different iterations.