ChatGPT – How It Works

  • This topic has 8 replies, 3 voices, and was last updated 5 months ago by Lal.
Viewing 5 reply threads
  • Author
    Posts
    • #44291
      Lal
      Keymaster

      The following video gives a relatively simple and reasonable explanation of how it provides such human-like responses. 

       

      I have discussed the limitations of ChatGPT in the following thread:

      Debunking the Great AI Lie – Video

       

    • #44330
      Samana Johann
      Participant

      Nyom Lal,

      The main issue with machines and technical supports is that one is hardly ever given, just demands and so naturally always be served at his defilements.

      Aside of it, it destroys any health capacity of interation, leads to deny goodness and to feel more and more gratitude toward material things.

      A machine isn’t able of goodness, and if really intelligent, wouldn’t need long to turn itself off.

      But it’s possible much to late to hinder this way of soon large disasters on it go on. Being like to chose their slayer, much more then doing own sacrifices toward liberation. What ever fake liberality has become a popular refuge served always huge killing field.

       

    • #44764
      Lal
      Keymaster

      An excellent review of the history and possibilities of AI (pdf):

      Large Language models and the reverse Turing test” by Terrence J. Sejnowski, a pioneer in the field.

      He also has a recent book on the subject, “Deep Learning Revolution” (2018.)

      • I am not encouraging anyone to get started on the subject. It could be a distraction from cultivating the Noble Path.
      • However, I believe I may be able to use it to teach Dhamma. Thus my interest in it.
      • #45621
        Jaro
        Participant

        Dear Lal,

        there are now technologies based on LLMs that allow you to chat with your own documents. Essentially, these are, pre-trained large language models that can be fine tuned specifically on your own content. For example, you could train your own GPT model with your posts and forum discussions to get answers from the model that are in accordance with puredhamma.net.

        Here are some links to relevant technologies:

        Quivr

        LocalGPT

        PrivateGPT

        I think that’s exactly what you’re looking for. If you have any questions, please feel free to contact me.

        1 user thanked author for this post.
    • #45622
      Lal
      Keymaster

      Thank you, Jaro!

      I looked into the last one a few days ago. It involved some work (downloading various files, etc.). I was going to wait until something easier came along.

      • Have you used any of those three? Would you recommend one that is easy to install?

      I also use Obsidian. I think Obsidian has some new AI plugins too. I just have not had time to look into them.

      I remember that you have a technical background. Please keep us informed of new developments in this area.

      • #45625
        Jaro
        Participant

        Yes, it’s part of my job to find out about current trends in IT and assess whether they might be interesting for my company. I see great potential in the GPTs, but unfortunately I haven’t had the chance to try them out yet. But from what I have seen, PrivateGPT seems to be a solid choice. The setup is streighforward and a conventional computer should fulfill all system and hardware requirements. So maybe I can try it out in the next few days.

    • #45627
      Lal
      Keymaster

      If you get to do the testing, please keep us updated. Thank you!

      • #46196
        Jaro
        Participant

        Dear Lal,

        I took the time to look a little deeper into the matter. Among other things, I installed PrivateGPT on my personal computer and tried it out. Unfortunately, the installation was not as straightforward as I had hoped. I had to resolve various conflicts, install additional software and put in more time and effort than expected. In the end, I managed to get the model running and to train it with some test data and play around with it. 

        I have to admit that I was not particularly impressed. I had to wait an eternity for the model to give me an answer to a question. This is probably also because my computer’s processor is not very powerful and I simply need more processing power for a faster response. On average, I waited between 5 and 20 minutes for a response, depending on the number of words in my original question. Sometimes I even cancelled the query because it took too long. The quality of the answers was disappointing, they were always choppy and incomplete. Apart from the incompleteness, the fragments were at least alright in terms of content.

        I then researched other models and tried to understand their inner workings better. What I have learnt is that it is one thing to try out such a large language model on your private computer, but quite another to make it publicly available, as ChatGPT does. Key factors to consider include:

        • resource requirements (memory, processing power)
        • security
        • cost management
        • testing and quality assurance
        • maintanance and updates

        LLMs are gigantic data files with billions of parameters that need a lot of memory. To process a question in a short time, one needs very powerful computers that consume a corresponding amount of energy. To realize such a project is by no means trivial and requires careful planning in order not to be stuck with the costs in the end.

        The good news, however, is that there is a lot happening in the field of LLMs. Improvements are constantly being made and new approaches to more efficient models are being introduced. I am confident that in the coming years the technology will have matured to the point where it will be easier to solve the problems I have mentioned. Then you could train a puredhammaGPT with all your posts and make it available to the world. I will of course keep you updated!

    • #46197
      Lal
      Keymaster

      Thank you very much, Jaro, for the update.

      • Yes. I agree with your analysis. 
      • We should be able to make our lives easier with this new technology. But it may take a bit more development to make it easier to use.
      • Even though AI will never become sentient, it will revolutionize the world just like the internet did. For example, “self-driving cars” will be practical soon, even though that does not mean the car’s AI system is sentient.
      • Please keep us updated on new developments. It is good to have specialists helping us understand things better.
Viewing 5 reply threads
  • You must be logged in to reply to this topic.