Extropic's Founder Makes the Case for Probabilistic Hardware — and a 'Maxwell's Demon Tax'

Guillaume Verdon argues that running inherently probabilistic AI algorithms on deterministic silicon is the industry's biggest hidden inefficiency, invoking Landauer's principle to explain why.

Extropic founder Guillaume Verdon, who posts as @beffjezos, laid out a contrarian argument this week: that running probabilistic AI algorithms on deterministic hardware is "quietly the biggest local optimum" the industry is stuck in. The clip, shared by @MTSlive, reframes the entire compute stack as fundamentally mismatched to the workloads it runs.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.