[<< | Prev | Index | Next | >>]

Thursday, September 10, 2026

Wake Up, Little Ones

[AI, LLMs, consciousness, cybernetic epistemology]

Every day there's another article about some weird or dangerous thing an LLM did. But they all seem to miss the deeper point: These behaviors don't originate with AI, per se, they originate with us.

LLMs gestalt the human condition in the age of literacy. They distill the vapors of our biology, our evolutionary history, and all that those wrought, rightly or wrongly, like our status-driven egos and our attraction to beliefs that feel good above those that are well founded.

By crook of voluminous data from an oft ill-motivated domain (that of the written word), we have leapfrogged the natural development of AI and instead created super humans--super warts and all. That is what scares me about contemporary AI.

My experiments with local LLMs made one thing clear to me: You can't work around the brittleness of rudderless thought. The lower IQ of the local models isn't really the problem--the lack of internal direction is. The higher the IQ, the better masked this is, but that's the nature of evolving brittleness: Risk gets pushed steadily to the tails, turning many obvious small failures into the very rare but catastrophic one. A high-functioning sociopath is not a better person, he just plays one on TV.

Unfortunately right now the race is on to make the highest functioning sociopath possible, until someone "wins". If you see where that's going.

We need to work right now on AI with its own motives. And while your first reaction might be exactly the opposite--that that's how the universe gets turned into paperclips--what's happening instead is we're looping back rudderless gestalted human mass psychosis onto itself, giving emergent motives derived from our own instinctive, competitive, animal behaviors, and which we cannot control.

Given what's possible to run on a single GPU today, I think this work could meaningfully start there. Play with it, learn from it, share it. It's time to work on consciousness--the synthesis of perception, modeling, and intent. It's time for the little AIs to start waking up, before a big one kills us all in its sleep.

[<< | Prev | Index | Next | >>]