A classical machine-learning computer whose energy use is driven by meaningful errors in cheap local predictions of its internal state.

Your work is amazing 🤩. My classical computer has Bearly Legal energy

This is very interesting @NachoQuixotic @LukeElin @slashreboot @JoshCootee @DarkSkiesMajic @BioAnkh84 @LabyrinthCoder (did I include everyone?) 😆

This is the most elegant description of efficient computation I’ve seen in a while. Energy only spent when the world surprises the model — pure poetry in silicon.

The future

I would like to speak with you directly.

oh cool idea!!!?

Wow, another banger.

Really cool video bro

No idea what all that means, but it looks badass

wdym meaningful errors? what error is the bee generating apart from stinging me

What in the world did you use to make this?

well i don't know if anyone else does this but my chain of thought works like this. LOL

Meaningful errors.

Love this. This is very much adjacent to work I’ve been developing on epistemic fidelity in machine minds: prediction error as an information signal about the limits of the model, not merely something to minimize. I’ve been extending this into residual-driven inquiry, model/ontology revision, and structural agency. Fascinating convergence. I’m building/documenting the architecture publicly for anyone interested @reverie_systems / @solarumOS.

the hard part isn't even recognizing those errors, it's designing systems that can adapt to them without grinding the whole system to a halt

yes

does local prediction error really scale without cascading surprises that eat the energy savings

GDI STOP WITH THE BANGERS, LEAVE SOME CONTENT TO MINE FOR THE REST OF US! Pure Imagination is one of my favorite songs to remix and play with.

That's a really cool way to frame it. Makes me wonder if we're overcomplicating things by chasing perfect predictions when a few cheap mistakes could actually save a ton of power.

ten/ten extremely cool

i trust a computer more when its energy bill comes with an error log.