• Re: Anthropic/Dept of War

    From Ron L.@1:120/616 to Aaron Thomas on Sun Mar 1 09:30:58 2026
    Aaron Thomas wrote to Dr. What <=-

    They work based on what they're trained on, but they also know how to
    be creative,

    That is a myth. What's really happening is that they are reasoning differently because they are unencumbered by human biases.

    Everyone has is biased. But most of us are unaware of it because the bias was "programmed" in by our parents and relatives and they all have the same biases.
    So it's "normal" for us.

    A while back, someone built an expert system (pre-LLM AI) for flavor pairings. All it was programmed with was "X tastes good with Y" type of information. But there were no "pickles taste funny" type of biases. It came up with flavor combinations that seemed strange to humans, but they were tried and they turned out very tasty.

    That wasn't creativity. That was clear reasoning without our biases.

    and they can figure things out faster than humans.

    Well, ya. Just like calculators can math far faster than humans.

    You're
    right that a commercial system shouldn't be used for defense. We need a truly MAGA AI model that will spit it out when we ask it how to defend ourselves in certain scenarios. Not have it say "The question you asked violates our community standards."

    I think that, right now, a truely generic AI is not reachable. AIs for specific tasks (i.e. programming) are certainly doable. But a generic AI will have to be free from biases and I don't see how that can be accomplished.


    ... I have but three enemies: fear, anger, ignorance.
    ___ MultiMail/Linux v0.52

    --- Mystic BBS/QWK v1.12 A47 2021/12/25 (Windows/32)
    * Origin: cold fusion - cfbbs.net - grand rapids, mi (1:120/616)
  • From Aaron Thomas@1:342/201 to Ron L. on Sun Mar 1 07:57:46 2026
    A while back, someone built an expert system (pre-LLM AI) for flavor pairings. All it was programmed with was "X tastes good with Y" type of information. But there were no "pickles taste funny" type of biases. It came up with flavor combinations that seemed strange to humans, but they were tried and they turned out very tasty.

    That wasn't creativity. That was clear reasoning without our biases.

    I get it; it's exploring options without bias, and it results in cool stuff that humans would unlikely ever think of (at least not quickly.)

    I think that, right now, a truely generic AI is not reachable. AIs for specific tasks (i.e. programming) are certainly doable. But a generic
    AI will have to be free from biases and I don't see how that can be accomplished.

    I'd like to give AI programming a try some day. There are frameworks that I can play around with. Google has one that's open source (TensorFlowPyTorch) but of course I'll look for alternatives to the root of all evil.

    --- Mystic BBS v1.12 A49 2023/04/30 (Windows/64)
    * Origin: JoesBBS.Com, Telnet:23 SSH:22 HTTP:80 (1:342/201)
  • From Ron L.@1:120/616 to Aaron Thomas on Mon Mar 2 07:06:32 2026
    Aaron Thomas wrote to Dr. What <=-

    I get it; it's exploring options without bias, and it results in cool stuff that humans would unlikely ever think of (at least not quickly.)

    Yes. As a software developer, I have to constantly check my biases when it comes to programming because sometimes I have a belief stuck in my head that steers me away from a good solution.

    I'd like to give AI programming a try some day. There are frameworks
    that I can play around with. Google has one that's open source (TensorFlowPyTorch) but of course I'll look for alternatives to the
    root of all evil.

    The concepts are easy. But it costs a lot in storage and CPU to make one that's fairly generic.


    ... if you want someone to keep a secret, keep it yourself.
    ___ MultiMail/Linux v0.52

    --- Mystic BBS/QWK v1.12 A47 2021/12/25 (Windows/32)
    * Origin: cold fusion - cfbbs.net - grand rapids, mi (1:120/616)
  • From Aaron Thomas@1:342/201 to Ron L. on Mon Mar 2 12:36:12 2026
    I'd like to give AI programming a try some day. There are frameworks that I can play around with. Google has one that's open source (TensorFlowPyTorch) but of course I'll look for alternatives to the root of all evil.

    The concepts are easy. But it costs a lot in storage and CPU to make one that's fairly generic.

    For my current projects (Node-JS non-AI stuff) I'm using a VPS with 16 cores and 16 GB RAM. Maybe if I migrate to something with twice the specs I could do it all on the same server.

    I'm always like 10+ years behind other developers, and this is no exception, but I really am excited to eventually jump into it.

    --- Mystic BBS v1.12 A49 2023/04/30 (Windows/64)
    * Origin: JoesBBS.Com, Telnet:23 SSH:22 HTTP:80 (1:342/201)
  • From Ron L.@1:120/616 to Aaron Thomas on Tue Mar 3 07:16:42 2026
    Aaron Thomas wrote to Dr. What <=-

    I'm always like 10+ years behind other developers, and this is no exception, but I really am excited to eventually jump into it.

    It's not called the "bleeding edge" for nothing.

    I'm usually "behind" too, but over 40 years, I've developed a good sense for what's going to stay and what's a fad.


    ... "Ground Beef" -- A Cow With No Legs!
    ___ MultiMail/Linux v0.52

    --- Mystic BBS/QWK v1.12 A47 2021/12/25 (Windows/32)
    * Origin: cold fusion - cfbbs.net - grand rapids, mi (1:120/616)