1d agoAnthropic's Akbir Khan argues massive AI speedup claims often stem from optimizing extremely inefficient, low-MFU baselinesGoing from 3% to 30% MFU yields a 10x speedupSentimentSentimentPos50%Neg50%Positive users call low MFU figures like 1-30% impressive for AI speedups while negative users sarcastically mock the claims by comparing them to a nanny's toe performance.4 comments with sentiment. View comments.