2h ago

Tom Cunningham posts chart comparing measured pass@1 scaling of general-purpose LLM on Erdős unit distance problem to implied best-of-N curve as accuracy climbs from 0.01 to 0.48

Noam Brown questions the base-10 assumption on the log axis.

——0——
Original post
OPNoam Brown#31Noam Brown@POLYNOAMIAL

@testingham You're assuming base 10?

2:43 PM · May 20, 2026 View on X
Tom Cunningham posts chart comparing measured pass@1 scaling of general-purpose LLM on Erdős unit distance problem to implied best-of-N curve as accuracy climbs from 0.01 to 0.48 · Digg