Embed Notice

Recent LLM advances immediately prompted warnings about the ability to •create• propaganda.

Perhaps less well appreciated is the threat of the model •training on• propaganda. These AIs have no model of meaning or correctness. All they can do is mindlessly repeat the patterns they hear — and what they “hear” is a random grab of whatever they can find on the web.

So if you pump obscure web sites full of disinformation, even if humans never see the sites…the LLMs will.
https://hachyderm.io/@cyberlyra/114127170072460743