Diarmuid O Keeffe ☘️

Models think, but algebra decides.

Economics takes a 4+ dimensional problem, models it for convenience as 3D (always missing the consequential dimensions), then analyses it as 2D while confusing the gullible public who interprets it as simple 1D.

@illscience - turns AI from stateless assistant into compounding operator - reduces repeated mistakes on cold start - creates an auditable control layer for regulated workflows - becomes switching-cost infrastructure: competitors can copy model calls, not your institutional decision memory

@FuzzieLabs Say more about trusted memory?

Agent networks open questions: - Do traditional network effects survive when participants are infinitely promiscuous? - Who owns discovery? Parallel / Exa? Google? Moltbook / agent native p2p network with something like DNS? - Do agent networks/machine networks have the same properties as traditional human networks (i.e., increasing returns to scale and an unassailable moat)? - What is even ownable? As we think about machine networks is there even a concept of proprietary supply or demand given that they can arbitrarily participate in millions of networks, join them, leave them, etc.? - Agents act on behalf of their human operators. Is there a concept of an agent as a semi-independent economic actor with a dependency on humans? - Stripe seems like the natural owner of the checkout experience, given that they have all the human payment credentials. They're also building the payments infra. Will they themselves try to become an aggregator of supply and demand and intermediate this network? - How should network operators think about agent acquisition, retention, churn, etc.? - What are the similarities and differences between these networks and web3 machine networks? - Who handles reputation / identity / fraud? - Will this time finally be different for micro payments on the internet?

@ivanburazin Claude CAD.

Ive been talking about this forever. About time. That being said, i think there is a dedicated CAD app/service for Agents to still be built

The masters are back... youtube.com/watch?v=xrZLtV…

Pope Leo has said that those who wage war are thieves stealing away our peaceful future, per Reuters

We are being deprived of the full transformative power of AI because none of the systems, not even @grok, have the power to say “WHAT THE FUCK DO YOU THINK YOU ARE DOING, YOU STUPID MONKEY?” Which is in fact the correct answer to many if not most questions posed to AI systems.

“If people cannot write well, they cannot think well, and if they cannot think well, others will do their thinking for them.” — George Orwell

The Kernel.

there is no version of me that will apologize for wanting to be exceptional. and I don't want to be close to anyone who needs me to.

If anyone w/ multiple GPUs asks me about the most important thing to learn about, it is Tensor Parallelism (TP) P.S. Buy your GPUs in 2^n, so 1, 2, 4, 8 and so on, otherwise TP doesn't work (mostly except for niche cases)

Album Cover:

trying to use topological data analysis to map the shape of my x bookmarks through mapper + embedding extraction and generated 3 views: - density: where attention keeps gravitating - pca: the dominant axes of variation - centroid: center vs edge (typical -> outlier)

Two major advances for Western science were: 1. The development of logic, and 2. the realization that we can discover causal relationships. Inspired by Turing, I have tried to put the two together... ~Conjecture Institute Advisor @yudapearl

@ashleyschendel As kids, we came back when the sun started going down. No one knew where we were. It was awesome.

You look at source code of best games ever made, and they are mostly for loops and if conditions over big fat structs. Really makes you think.

This megakernel is using a 3090. Stock tinygrad beats this (420 tok/sec!) using a cheaper 7900XTX. With our custom driver AMD hardware can really shine.

This is how Hermes agent creates skills and develops memories btw