What you described about the use of random numbers by neural networks is incorrect.
if you actually want to understand how they work, you might start by Googling "tom ai beginners" without the quotes. my article is usually the first hit in the list that google returns after you've waded past the paid for stuff and the video links. After you've understood the contents of that piece, you might then move on to some very excellent stuff which you can find by googling "victor zhou neural networks” again without the quotes. unlike most of the drivel about artificial intelligence and how it works, Victor gives actual examples of different kinds of neural networks, including the code and meticulous explanation as to how the code works. On Tue, Mar 17, 2026 at 12:06 AM <[email protected]> wrote: > > (please don't top post: it confuses the hell out of everyone) > > > On Sat, Mar 14, 2026, 7:02 PM Dan Ritter <[email protected]> wrote: > > [...] > > > > Yes, I too have always found that rolling random numbers and > > > hoping that they don't lie to me is an effective way of running > > > complex systems. > > On Mon, Mar 16, 2026 at 04:50:50PM -0700, tom arnall wrote: > > Rolling random numbers. To what are you referring with this term? > > Generative models (and that's what those chatbots and image generators > currently being marketed are) basically run a random number generator > to pic between nearly identically probable choices. > > Very roughly put, they feed a statistical pattern recognizer with > white noise, and the most probable recognized patterns become its > output. > > Then, of course, there is a lot of manual work going into avoiding > problematic behaviour (giving wrong medical advice, illegaly giving > legal advice, yadda, yadda: whatever we, in the "first world" consider > yucky). This work is mostly done in low income places (Nairobi, f. ex.). > > Cheers > -- > t >
