Comment on DeepSeek: how a small Chinese AI company is shaking up US tech heavyweights.

nesc@lemmy.cafe ⁨3⁩ ⁨days⁩ ago

• Cat@ponder.cat ⁨3⁩ ⁨days⁩ ago

  Obviously I am hyped about this. With that in mind:

  fake benchmarks

  Any proof fo this?

  censored llm

  Is ChatGPT uncensored?

  source

  • nesc@lemmy.cafe ⁨3⁩ ⁨days⁩ ago

    No chatgpt is censored (why is it even relevant, lol), and they censored it so well that it became pretty useless last time I’ve used it. But even compared to other chinese llms deepseek is censored too much.

    No, I don’t have a proof, I’ve tested it a bit and replies were full of factual errors, it tried to gaslight me multiple times, in general it behaves closer to something like llama 8b from two years ago. I haven’t tried the “reasoning” model yet, maybe it’s really great (openai one was shit, so no high expectations here).

    source

    • Cat@ponder.cat ⁨3⁩ ⁨days⁩ ago

      After understanding your opinion, your comment seem to be reasonable but was put with bad presentation, imo.

      I think being cautious might be a good point here. We are very early here and we will definitely see a better benchmarks for this model as the time pass. Which will prove or disapprove their benchmarks.

      source

      • nesc@lemmy.cafe ⁨2⁩ ⁨days⁩ ago

        Yeah, sorry, as time goes on my english gets worse. 😺

        source

        • -> View More Comments
  • Havald@lemmy.world ⁨3⁩ ⁨days⁩ ago

    Is ChatGPT uncensored?

    Does it have to be for this to be a valid criticism?

    source

    • Cat@ponder.cat ⁨3⁩ ⁨days⁩ ago

      The title is “how a small Chinese AI company is shaking up US tech heavyweights.”

      I am not saying that this criticism is not valid, what I was arguing for here is that the best USA LLM is also censored in comparsion, so the fact that we got something opensource that we can uncensor is better than ChatGPT closed and censored model.

      As far as I know, the only used LLM family that is uncensored and can function locally is Mistral AI models.

      To put it simply, DeepSeek could be decensored. But it will take time to do so.

      source

      • BaroqueInMind@lemmy.one ⁨2⁩ ⁨days⁩ ago

        Apparently, if someone had the resources to get a hold of some decent nvidia A6000 gpus, they can abliterate the LLM in six hours.

        source