@edsanders2 @cwebber Agreed LLMs wont do anything not told to do, but they are kind of grown.

This is not the greatest article but as I understand it they are made like: https://blog.ciaops.com/2026/04/12/llms-are-grown-not-coded-and-that-changes-everything/

This book goes into more details about how weird LLMs really are and then falls into doom but the authors are respected AI specialists and they say pretty much the same thing: https://en.wikipedia.org/wiki/If_Anyone_Builds_It,_Everyone_Dies

@cwebber How do we "slow" down development of LLM's if they are grown rather than programmed by a team of humans? Would that even be possible?

I think its also important that Deepseek released v4.1 flash just before the weekend with tech (most of the model is on normal ram or even ssd and 8gb in vram) and at price points that no US company can even hope to match currently. I would be shocked if the teams at Kimi and Z.ai haven't been breeding from that model greedily over the last few days.