When presented with conservative-biased chatbots, Democratic voters supported substantially reducing government expenditures on education, while Republicans did the opposite when presented with a liberal-biased chatbot.

you are viewing a single comment's thread
view the rest of the comments

[–] gay_king_prince_charles@hexbear.net 8 points 6 months ago (2 children)

Ironically, if you're using a prompt and Marxist-only RAG, you'd have to use a Western LLM for that. Deepseek's censors are ratcheted down to the point where it becomes useless for anything political as it errors out whenever it mentions Mao, Deng, or Xi (although it can mention Zhou Enlai without issue). Kimi or Qwen might be a bit more permissive, but I believe they still crash upon mentioning Xi which is a massive downside when trying to make a Marxist LLM.

[–] MayoPete@hexbear.net 6 points 6 months ago (1 children)

If this were to take off I imagine it would have to be a Llama trained homebrew LLM and not anything owned by a corporation. I imagine pointing an LLM at marxists.org and transcripts of Matt Christman's videos and go from there.

[–] gay_king_prince_charles@hexbear.net 5 points 6 months ago (1 children)

Llama is owned by Facebook, that's what I mean by Western

[–] MayoPete@hexbear.net 5 points 6 months ago

Ah didn't know that

[–] AtmosphericRiversCuomo@hexbear.net 1 points 6 months ago (1 children)

I thought the API isn't like that, only if you use it through the UI. Or you could just use another version of deepseek hosted elsewhere.

[–] gay_king_prince_charles@hexbear.net 1 points 6 months ago

The API and web UI have the same censors. I believe they are baked into the model itself.