It attributes stagnant AI benchmarks to a fictional sophon block.

@xlr8harder I’m biased bc I’m anti Anthropic / pro OpenAI but I doubt they use test time, maybe they just omitted stuff from training

@xlr8harder hehe I thought of this instantly asw
oh no

@xlr8harder This explains a lot about the trend being vaguely down
It attributes stagnant AI benchmarks to a fictional sophon block.