Sam Hartman <hartm...@debian.org> writes: >>>>>> "Simon" == Simon Josefsson <si...@josefsson.org> writes: > > firmware blob > Simon> for a future SoC CPU that includes camera functionality, it > Simon> seems possible that would make use of some LLM model to have > Simon> better face recognition for example. > > Do you perhaps mean model or machine learning model rather than LLM? > I think that LLMs are large enough (even the "small ones") that we'd be > aware if they were in non-free-firmware today, and at least the tasks > you are talking about sound like they would be better approached by > machine learning rather than something specifically directed at natural > language.
Yes, sorry, my terminology is sloppy and I tend to conceptually merge all these. From my point I view I don't see them as necessarily any different from a include-in-Debian-or-not point of view. Is there a significant difference between any of these terms for this discussion? Size of the model is the only one I can guess, but I also believe that there are LLM's smaller than some of the bigger machine learning models, so I don't think it is that relevant. Having a small LLM model in a non-free firmware blob to bootstrap a text-to-speach or speach-to-text input method in a laptop doesn't seem far fetched to me. /Simon
signature.asc
Description: PGP signature
