2026-10-04 07:43 UTC
Replies (1)
-
@bnuuy@mathstodon.xyz 2026-10-04 11:14
@creepy_owlet@mastodon.online hm, i do have to point out that "embraces LLMs", in this case, appears to mean, "using LLMs for the incredibly tedious task of reverse-engineering undocumented proprietary hardware", not necessarily for writing any actual code. this is maybe one of the few uses of these things i can tentatively get behind: this is genuinely gruelling and unnecessary busywork (apple knows how their hardware works, but won't tell us), it benefits the public to create public documentation of this hardware, and it in the long run fights back against the planned obsolesence of hardware ("ah shucks, the manufacturer stopped providing updates, and nobody else knows how to program this bespoke hardware, i guess it'll have to go in a landfill now"), povided the documentation is done well. and since everybody else has to deal with LLMs being able to dig up security vulnerabilities en masse that were previously too tedious to find/tedious enough that one person finding it could reasonable assume it was secret for the moment, maybe it's fine for the tech industry to at least get a little taste of that too, once in a while. the project page is a bit unclear about where exactly they stand on LLM code generation though, since they are very focused on the part where they want to use it for reverse engineering, and very heavily insist on air-tight clean-room practices. i'd really still prefer them to write their own drivers etc by hand, but i would have a bigger bone to pick with the general linux kernel first 🙄