CNN just did a piece on private Israeli groups doing the exact same thing. It is pretty scary the shear scale of this. Literally any entity or group can spin up a ton of bots and use any AI service locally or otherwise, to attempt to sway public opinion.
Which… actually that’s not true. Iran isn’t using pencil and paper to influence US elections, they’re using LLMs. And if all the big cloud providers cut them off, they’ll move to using home brew models. That is the impact and putting our head in the sand about it isn’t going to be helpful.
This is NOT to say we should not release to models! I think we should release open models, and that in addition to that we should also care about written impact assessments so everyone is on the same page about risks! I think it’s great that Jeremy Howard is releasing open foundation models, but I wish he wasn’t so flippant about the real world implications. I’ve always looked up to him for the work he has done, and found that attitude very disappointing.