John

The level of access I have to information through LLMs would be inconceivable to my ancestors. Not impressive. Not convenient. Inconceivable. For most of human history, knowledge was locked behind distance, class, language, institutions, memory, and luck. Now I can ask a machine to cross-map fields before breakfast. The danger is thinking access means wisdom. It doesn’t.

@AndrewHires @julianboolean_ @Loh Great corrections. Where do you get spikes/s data from?

If you expand the definition to encompass motor signals etc to the muscles of the eyes, (which is admittedly how the question was written), feed forward connections still outnumber feedback like 50:1. If you do it in spikes/s, then feed forward still wins by >10:1, except perhaps for a few ms during a saccade.

damn, TIL the brain sends more info to the eyes than the eyes send to the brain

@micahgallen @vineettiruvadi Sequencing or release shaping is a better compression for what the cerebellum does.

The cerebellum traditionally governs motor control, yet emerging research reveals its circuitry is also active during working memory, language, and social cognition tasks. journals.plos.org/plosbiology/ar…

@Qivshi1 Say more

Once you start looking for the oscillations and waves...

@societyhatestee Character

what's it called when you have enough trauma to be evil but actively choose not to.

@PostLinguistic Bro can navigate through the void

@ArticlesOnX Doing my best.

I feel free to speak my mind on X as no one will ever read it.

Fair, but I don’t think you need to zoom in that far. A lot of the “molecule and gene” activity has to do with cellular burden, important for pathology discussions, not so much for AI computational improvement ideas. Signaling, receptors, local compounds and the conditions they create might be interesting for future AI improvements, but that is another level of advancement. Routing (circuit motifs) between layers is what I am recommending. Motif deployment in a GNN should reduce hallucination and resource requirements.

Agreed. Those brain circuit patterns could someday really help fix the info loss in GNN message passing and keep better track of context. The real difficulty is that brains rely on a precise mix of boosting and dampening signals, learning rules that adjust locally, and sparse efficient connections. None of that maps neatly onto the dense, heavy math we use to train networks on GPUs. Adding true feedback loops or inhibitory parts often makes training unstable or demands far too much computing power on biological graphs like molecules or genes.Its a promising direction, but its a nontrivial architectural leap beyond current attention tricks like GAT.

Graph neural networks, in theory, should be far more useful for AI in biology than transformers or LLMs. Biological systems are naturally organised as high-dimensional, non-Euclidean graphs - molecular graphs, gene interaction networks, knowledge graphs, etc. Yet transformers still dominate GNNs on tasks like molecular bioactivity prediction - even when GNNs are enriched with physical properties about atoms and bonds. The core problems are information loss during message passing and global pooling layers. Adding attention mechanisms helps (GATs, GTNs, global attention pooling, etc.), but it’s not enough. What the GNN field and AI/bio desperately needs is a revolutionary new architecture - an "Attention is All You Need" moment for GNNs, as significant as the transformer.

@vineettiruvadi Could you provide some examples of what you see as redundant subsystems?

@PostLinguistic Many subsystems have massive redundancies yes. Unclear how much redundancy emotion/regulation and executive function have atm I think...

Not a dunk, not trivializing, but one of the most important near-facts I learned in medicine is that most 60yo+ men have swiss cheese lacunes because of countless "silent" strokes.

@mudscryer Say more!

It looks like I finally genetically modified a fungus by myself for the first time… MYCOENGINEER OFFICIALLLL

@OGdukeneurosurg Left and right eye field processing is something else.

How vision works: an anatomical view of pathway.

@_The_Prophet__ It’s uncomfortable.

⚡️To hear what is coming, something false in you must die first.

@kanair Say more

There seems to be a strong cognitive bias that makes us believe everyone else is wrong about consciousness.

@_The_Prophet__ Banger tweet.

⚡️The false thing needs a thousand voices to remain alive. The true thing can survive a thousand years without one.

@ravishar313 No lies detected

Biology will soon be an engineering subdomain

@DabsMalone Never split the difference

» The Alchemist » How to Win Friends & Influence People » 7 Habits of Highly Effective People » Sapiens » Autobiography of Benjamin Franklin » The Richest Man in Babylon » Almanack of Naval Ravikant What am I missing?

@OGdukeneurosurg This is great, Thanks for sharing!

Brain anatomy