Loading...

KK7NQN. Alright, Mal, looks like we're going to get an expert weigh in on this. KK7NQN, what do you got? So yeah, on the AI, the LLMs are right. They don't exactly learn after you've done initial training. The models are on a fixed training, but they do kind of adapt with context. So as the thread continues, they can increase their understanding of the current context that is being passed. So that's why I do modules that have their own individual tasks that they do. The other thing that I do is that way the stored context of the data going into the expected output stays consistent. As far as data sharing, it's kind of difficult because of the different layers of it. The model has its fixed data and the contact is their thread, and when you mix it you get inconsistent outputs. And computers are not a fan of inconsistent outputs. The thing I do to kind of share it is there's a set of tables that has its own table, and each step can look backwards. But the data is never modified in its original place. It is read from one place, modified, put into a new place, and just moves down the chain. So the original data is never malformed. Because again, we need to maintain our context of that thread to make sure that our output is exactly what our scripts and code expect to read. Because again, computers don't have awareness like what a human would have to adapt to different data structures. So yeah, AI with human speech is actually really difficult because slight variances in how stuff is said can cause big problems. Like I was saying, one of the things that I always fight with right now is just slight mishears. Like PSRG and Puget Sound Repeater Group are identified as two separate clubs because the computer hasn't associated that yet. I have to manually write something that does that association, and I just haven't yet. Anyways, KK7, NQN, back to you.