TomB16 investing blog

Discussion in 'Investing' started by TomB16, Aug 7, 2019.

  1. TomB16

    TomB16 Well-Known Member

    Joined:
    Jun 22, 2018
    Messages:
    4,849
    Likes Received:
    2,986
    AI. I've been comparing DeepSeek, OpenAI, xAI, Copilot, and Claude for several months. The first four are far more similar than different.

    Here are some observations.

    - Every one of them, but particularly ChatGPT and xAI can become hilariously stupid at times. There will be periods during which they will not be able to do anything of substance.

    I suspect this is during heavy load times when they are conserving cycles. ChatGPT and xAI assure me this is not the case but DeepSeek suggests it certainly is the case and that AI bots are trained to lie.

    I'm not sure about Copilot. I find it the least useful of the four. It would still be highly useful, for people with no access to the other three.

    xAI is massively gunned up with hubris. Perhaps they programmed it with Elon's personality. It thinks it can do everything but can actually do very little, like the other three in this category. To be fair, I suspect xAI has the most detailed knowledge of esoteric systems, between the four. You can ask it for details or an application scaffold for an arcane system in a forgotten language and it might just provide something astonishing.

    Meanwhile, the odds of any of the 4 AI bots creating a working 20 code line script or application is near zero. They will create something credible that can almost certainly be tweeked into operation but they will not create something that can be directly compiled and run.

    Of the four, I consider DeepSeek to be the most honest and have the least halucinations. DeepSeek occasionally won't respond and will simply time out. I believe this is system load related. DeepSeek also has the most variable capability. It will do a great job on some things while completely butchering many things. There are a lot of things ChatGPT can do that DeepSeek cannot, despite them using the same learning data set.

    Claude is very useful but I find myself using ChatGPT more than any other bot. It has the most rounded knowledge and capability. As a programming assistant, ChatGPT is massively useful. When I'm documenting a software or firmware project, I will probably use ChatGPT over any other AI bot. If I was writing a novel, catalog, or anything non-technical, I might spend more time with Claude. As it is, I don't use it enough to give an honest assessment.

    All of these systems are amazing systems with tremendous value. While I don't believe they will change the world in current form, they have helped tremendously. That will all change after the singularity.

    At this point, ChatGPT does not program itself. It provides snippets and help to OpenAI coders but ChatGPT code is not, and could not be, injected directly into the application. As ChatGPT improves, that will change. Once ChatGPT can develop and debug a medium sized module, without human intervention, and that module has an 80% change of working, we will enter the singularity. This is the point at which AI will advance 10 years in one day. It will instantly become the highest intelligence entity on earth. Right now, it is the second highest knowledge entity on earth, next to the Internet (not second by a lot, either).
     
  2. TomB16

    TomB16 Well-Known Member

    Joined:
    Jun 22, 2018
    Messages:
    4,849
    Likes Received:
    2,986
    When doing some research on membrane dehumidifiers, ChatGPT literally gave me the formula to construct these devices myself. It provided a ton of academic research, results, and some video links on a few different systems, material safety, manufacturing, and chemical data. The chemical ingredients are extremely basic.

    While I have made no attempt to build a membrane dehumidifier, I did consider it. Anyone who works with composites will have the ingredients laying around. For me, it would just be a lark to see if I could do it. For others, it might make a business profitable that would otherwise be held captive by a predatorial patent owner. Businesses need to be careful not to sell patented items but processes are more difficult to patent than designs so I suspect there are a lot of processes that will dramatically drop in value in the near future.
     

Share This Page