Vladimir_Nesov

Quadrillion Param Costs: KV Cache, Con­text Length, Fron­tier Margins

Model Size Scal­ing in 2023-2031

Hu­man Agency in a Su­per­in­tel­li­gent World

Mus­ings on Re­ported Cost of Com­pute (Oct 2025)

Per­ma­nent Disem­pow­er­ment is the Baseline

Low P(x-risk) as the Bailey for Low P(doom)

Mus­ings on AI Com­pa­nies of 2025-2026 (Jun 2025)

Levels of Doom: Eu­topia, Disem­pow­er­ment, Extinction

Slow­down After 2028: Com­pute, RLVR Uncer­tainty, MoE Data Wall

Short Timelines Don’t De­value Long Hori­zon Research

Tech­ni­cal Claims

What o3 Be­comes by 2028

Mus­ings on Text Data Wall (Oct 2024)

Vladimir_Nesov’s Shortform

Su­per­in­tel­li­gence Can’t Solve the Prob­lem of De­cid­ing What You’ll Do

OpenAI o1, Llama 4, and AlphaZero of LLMs

Mus­ings on LLM Scale (Jul 2024)

No An­thropic Evidence

A Math­e­mat­i­cal Ex­pla­na­tion of Why Char­ity Dona­tions Shouldn’t Be Diversified

Con­se­quen­tial­ist For­mal Systems