@itzzen@solarpunk.moe @modulux While that is true for the "littering the public sphere with crap" aspect of it, the "worker exploitation" aspect is hairier - even when building something for a nominally legitimate purpose (which accessibility definitely is, if done right!) it still matters exactly how the training data was sourced.

And while there are plenty of things called "AI" that can pass that test, for LLM stuff in particular it's not clear that it's even *possible* to do with ethical sourcing due to the sheer quantity of training data needed.

Mozilla has, crucially, been very quiet on that point of where they intend to source the training data from.