💾 Archived View for kotobank.ch › ~merlin › llm_censorship.gmi captured on 2023-09-28 at 15:52:03. Gemini links have been rewritten to link to archived content

-=-=-=-=-=-=-

LLM and Censorship

I had a thought that seemed trivial, but for some reason, I don't see it in my media bubble.

New models make it easy to check texts for compliance with censorship criteria. Human censors are slow, can go crazy, and may have their own agendas. LLM only needs hardware, and for new open-source models, it doesn't even have to be the most powerful.

Even the most fastidious censors can write prompts, a task that any bureaucrat can handle.

For the next six months to a year, the internet will save lawmakers and bureaucrats from their clumsiness. But I'm sure the gears are already turning.

In the next five years, we can expect a boom in near-real-time censorship of messages on social networks and messengers. This will be presented under different sauces - the fight against foreign influence, hate speech, or pedophiles. The process will affect both democratic and, obviously, autocratic countries.

Combine this with the gradual squeezing out of end-to-end encryption and decentralized social networks, and you get a rather unsightly picture.