Sunday, November 01, 2026

DIY LLM

Next Token Prediction with Neural Nets

We suppose it unlikely that the Valley Bros will embrace this analogy (LLM = Ouija board). For one thing, it doesn't clearly separate training from performance in the field, as the neural nets we connect to the planchette (cursor) are already worldly (trained).

Fortunately Silicon Valley and its captured investors + feeder universities are not fully in charge of the PR. If we in Silicon Forest wanna teach our young about hyperstitious beliefs outta California, we’re free to.