Expertium

10x more train­ing com­pute = 5x greater task length (kind of)

Let’s look at an­other “LLMs lack true un­der­stand­ing” paper

Smarter Models Lie Less

In­tel­li­gence Is Not Magic, But Your Thresh­old For “Magic” Is Pretty Low

Let’s stop mak­ing “In­tel­li­gence scale” graphs with hu­mans and AI

Ques­tions for old LW mem­bers: how have dis­cus­sions about AI changed com­pared to 10+ years ago?