Ryan McCorvie

probability, backgammon, sumo. Based in Oakland / Berkeley

[Ryan McCorvie @mccorvie]

Dario Amodei said if he invests $1T in compute but demand is only $800B then "there’s no force on earth, there’s no hedge on earth that could stop me from going bankrupt". The central tension for AI economics is whether capabilities and revenues arrive before the bill for compute comes due.

I created a toy model which casts AI economics a ruin problem.

- whether AI is a bubble comes down to a single parameter q
- high capitalization and intense competition leads to a race where no one can buy an edge

https://t.co/XK6WHNY2IA

[Ryan McCorvie @mccorvie]

@justjoshinyou13 I was trying to get color on exactly this question, to what extent are these breakthroughs constrained constructions

[Ryan McCorvie @mccorvie]

its like falling through the event horizon. Locally nothing happens, but to distant observer you redshift and hang forever

[roon @tszzl]

passing right through the superintelligence boundary

[Ryan McCorvie @mccorvie]

There should be a "cascade theory of markets" where leverage and the pain trade are the central features. Alternative to EMH / random walk. An endogenous explanation for heteroskedasticity, excess volatility, and crashes

[Ryan McCorvie @mccorvie]

Situational Awareness succumbed to the pain trade. Speculators move the market against leveraged investors, investors stop losses, the market moves more, rinse repeat.

[Ryan McCorvie @mccorvie]

mathematics in the age of mechanical coproduction

[Sebastien Bubeck @SebastienBubeck]

yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model.

We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more.

More thoughts here: https://t.co/8SjXONeh38

[Junyan Xu @Junyan_Xu]

1/6 Classification of compact surfaces, solved by Ryan McCorvie @mccorvie (collaborative work of the SF LEAN meetup) and Jack McCarthy with mostly ChatGPT 5.6 Sol on Jul 30, 2026, in 141k lines of Lean.

[Ryan McCorvie @mccorvie]

@jasondeanlee nah the server was down

[Ryan McCorvie @mccorvie]

Imagine how much we'd understand about language if Gauss had stuck with philology

[Ryan McCorvie @mccorvie]

@dylgr0 very cool!

[Ryan McCorvie @mccorvie]

cthulu x-risk: the possibility that AI's, just by correlating knowledge across previously disparate domains, will unleash some terrible info hazard

[Scry Visuals @Scrygl]

I asked Claude to hunt for a genuinely new math result. We chose cellular automata. It wrote its own search software, and found a spaceship nobody has ever recorded — in a universe one rule away from Conway's Game of Life. 🧵

[Ryan McCorvie @mccorvie]

@Aron_Adler @slimjimmy #TeamLeanStream

[Ryan McCorvie @mccorvie]

@HedgieMarkets It’s a failure of copyright law not the malice of tech companies . Google wanted to digitize every book with no clear business model or use for the data but were stopped. So this destructive corner case is all that’s lef

[Ryan McCorvie @mccorvie]

full sumo elo-tron here

https://t.co/JVOJG9Prt7

[Ryan McCorvie @mccorvie]

Aonishiki *and* Kirishima now have higher elo than either yokozuna. Tough tournament for the yokozuna.

[Ryan McCorvie @mccorvie]

@matthew_pines Sometimes I just threaten to increase the reasoning level. You know to motivate them

[Ryan McCorvie @mccorvie]

Full elo-tron here

https://t.co/JVOJG9Prt7

[Ryan McCorvie @mccorvie]

Kirishima now has higher elo than either yokozuna

[Ryan McCorvie @mccorvie]

Straightforwardly true

Geometry repurposed the visual cortex, algebra the linguistic cortex, real analysis the cerebellum (apply the triangle inequality by rote), and category theory the amygdala (it’s scary)

[Tero Parviainen @teropa]

Terence Tao says evolution hasn't given us a "math center" in the brain, so different people repurpose different areas of the brain to solve math problems.

Which leads to different styles of learning and communication.

[Ryan McCorvie @mccorvie]

@op_idempotent very slick... approx comes from Laplace's method or CLT on sum of exponentials (ie gamma dist)