Extropic's beffjezos: everyone optimizes intelligence per dollar, not raw intelligence per second
beffjezos · x · 2026-09-23
Extropic founder beffjezos argues the industry over-optimizes "intelligence per dollar" and under-optimizes "raw intelligence per second." Quoting a hands-on comparison where Astra X High Fast simply beats Opus 5.5 Max on speed, he calls Astra's fast mode a genuine moat and predicts OpenAI's strategic fast-inference partnerships with Cerebras and Jalapeno will become huge moats.
The thesis: once frontier model capabilities converge, inference speed itself may be the next differentiation battleground, with dedicated inference hardware partnerships racing to claim it.
More from AGI Musings
- NYT Editorial: Humanity Restrained Nuclear Tech Once, and AI Can Be Restrained Too — connoraxiotes · 2026-09-23
- Frontier Math Will Need Billions Like Big Physics, Say Researchers on 'Big Math' Era — sytelus · 2026-09-23
- An AI prompt workflow for equity research: bear cases, filings, and management promises — thisguyknowsai · 2026-09-23
- Meiji-scale AI disruption demands real 'opt-out' options, researcher warns — dioscuri · 2026-09-23
- AI cracks 100+ open math problems but general-purpose physical AI remains missing — beffjezos · 2026-09-23
- Ben Todd calls out OpenAI's circular logic: automating AI research to defend against the superhacking AI it creates — ben_j_todd · 2026-09-23