• 0 Posts
  • 302 Comments
Joined 3 years ago
cake
Cake day: July 2nd, 2023

help-circle





  • To my surprise almost if not all LLM are set not to be deterministic and have one unique input result in always the same output.

    They all are set to have a “temperature” setting so that they sound more natural.

    Personally I think quality of the output tokens of LLM is surprisingly not as much their priority as the quality and truthfulness of the result.

    I would 100% prefer a LLM that is purely deterministic and repeats the same answer exactly to the same question. Instead LLM are constantly choosing the next likely token more TL appear human rather than being accurate.

    These LLM are designed as sycophants and set up and trained as such.

    So a fingerprinting in the output seems quite realistic. An LLM is not giving you it’s best most likely answer. It’s taking one of the most likely answer and adds a sprinkle of uncertainty and randomness on top of it so it looks natural…


  • I applaud the open source Chinese makers. They might be “distilling” from the closed weights models of competitors but since they themselves pillage the internet with no oversight in my book that’s fine. In some way the Deepseek, minimax and kimi makers are forcing the weights of bigger proprietary open.

    Right now on OpenRouter a large part of the processing goes to Deepseek. Deepseek is still weaker than the Claude “frontier” models but they are way cheaper. Since they are open weights anybody can compete as providers and the cost stays “low”. These AI Chinese makers are a constant threat to the american proprietary models and that’s a great thing for the consumers.








  • Wow

    My primary concern is the risk that authoritarian governments—not solely the Chinese Communist Party (CCP), although the CCP is clearly the most capable threat—build AI models that are more powerful than those built by the US, and use them to achieve permanent military superiority or perpetrate incredibly deep repression of their own people.

    What is that shit?

    Truly this is an insane take coming from an American company. Really batshit crazy.

    Imagine the dread for Europeans to see these powerful countries obsessed at weaponizing their AI. Open source model are not even part of that problem. Open source is usually the opposite end of the secrecy of weapon design and manufacturing.

    But no evil CCP bad can’t do their open weights AI but OpenAI can equip the military without any oversight.

    I hope the American imperium will crumble and fall. They created so much fucking suffering for the rest of the world.







  • Tetsuo@jlai.lutoLinux@lemmy.ml•LTT does another Linux Challenge
    link
    fedilink
    arrow-up
    21
    arrow-down
    1
    ·
    8 months ago

    I did the switch to full time linux gaming a few weeks ago.

    Initially I was on PoP OS but I wasn’t happy with it. I reinstalled everything on Cachyos and it has been very good so far.

    Here are the main “hurdles” that I think I should not have encountered in my “liberation” :

    • I activated FDE in cachyos but the LUKS decryption prompt was never on my keyboard layout so I could never decrypt with the original key. I should not have to handle keyboard layout issues when decrypting FDE. This would be fatal to any beginner.
    • I still think the tinkering with proton is too involved. Some command line arguments should be activated automatically depending on your setup. Players should be able to anonymously opt-in to sharing settings for all the game they start.
    • Mouse sensitivity is wonky on many linux distros for gaming. Mouse acceleration should not be a default if you target gamers. I dont care if original Arch has this default. 99% of FPS players avoid it.

    That being said, leaving Windows makes me incredibly happy and I’m very thankful for the great opportunity the open source community has given me through great free software.