As I wait for death, or the next inevitable inversion of reality, I tend to tinker, fiddle and play.
I am currently sidetracked into this little realm of artificial intelligence, and machine learning. Pretty fascinating stuff, if also a little too on-the-nose for trying to build our own version of fuzzy logic using hardwired electronics and semiconductors.
So at once I both question the usefulness of building a computer into a fallible weighted reasoning mechanism and recognize the utility of not having to learn how to do *everything* by myself, it’s fun and interesting to tinker with the possibilities.
And when we ask AI for an answer to our question, it’s going to have access to the best, closest-to-what-we-actually wanted answer, trained from a real-world dataset created when expert X performed experiment Y and got Z. Unfortunately, it will also have the option to choose from among ALL the rest of the uninformed, ignorant, trolling, nonsense that anyone else might provide as the answer to your question. And for the most part, the LLMs are not effectively evaluated for things like truth or actual results. They’re just trained on a lot of people writing a lot about a lot … whether the content of the writing is correct or not.
But hey, I certainly see how folks are excited about the possibilities, and I can only wonder how many people have gladly supplied the owners with … all their thoughts and feels. Who knows what’s next.
I SUPPOSE, that I should move the ANN off the Arduino and onto it’s own processor, and just let the Arduino handle all the main management tasks of routing signals around for processing by different bits.
I think what I’ m going to end up wanting to do is train different portions of different ANNs for different functions, rather than trying to build them altogether as one. Perhaps there’s one ‘main’ ANN for personality and LLM functionality, and a separate, more purpose-oriented model running on it’s own to handle other tasks.
I’m sure I’m pondering things that others who’ve already immersed themselves in the field, or who are at the cutting edge of development are past, but it also amuses me how many times I’ve thought that certainly this simple concept is understood already.
Anyway …. I don’t have any real friends, so I’ll get back to trying to design some.