Meta AI Research – An AI Model of the Human Brain
aidemos.atmeta.comTRIBE v2 (Trimodal Brain Encoder), a foundation model trained to predict how the human brain responds to almost any sight or sound.
I helped implement the demo
This is cool. I envy you for being able to work on projects like this.
There's a spectrum of intellectual ability, and I tend to find myself interested in things far lower on the scale - insects, worms, amoebas, etc. I have a strong interest in GOFAI - pathfinding, goal-oriented action planning, expert systems, etc, and have played a lot over the past decade or two with implementing these things in roguelikes - simple 2D-ish world, sensory stimuli, etc.
So the recent stories about teaching a wad of human neurons to play _DOOM_ and this specific model are pretty interesting to me, even if, again, it's a few orders of magnitude in complexity higher than my focal point.
Are you aware of any similar work that's going on at lower levels with simulations like this? Did TRIBE v1 follow from simpler models or was it "greenfield", so to speak?
(Apologies... I figure I could Google my questions but sometimes I learn more from an HN comment than I would in hours of trying to parse a graph of scientific papers and news articles.)
Thanks for the kind words. I helped on the brain rendering side of things but I’m not an expert in the field he he. You seem to be way more knowledgeable than me. Do you have any of your work out there? I would love to check it out.
I don't know anything about anything TBH. I'm just an interested dude. My day job is just working with microservices and stuff, nothing related to AI or ALife.
Probably the closest thing I have right now is a p5js sketch with a genetic algorithm here: https://bitterbridge.github.io/p5js-sketches/robbie-the-robo... (takes a few minutes to start working effectively and probably has a ton of bugs).
I've started making some other sketches like https://bitterbridge.github.io/p5js-sketches/monsters/ to set up this kind of roguelike idea for demoing some more of these things, like Goal-Oriented Action Planning, but I haven't made much progress yet.
I also have this Lisp rules engine VM I created over the winter break, https://github.com/ndouglas/longtable/ , which might become the core of a more complex reasoning system. IDK though.
So little time, so much to do :)