Gijs Verheijke

 

Unpacking Deepseek

28 January 2025 Medium: Substack

Unpacking Deepseek by Gijs Verheijke

Benchmarking AI is a hard problem. And as the benchmarks become the goal (Goodhart's law), the model makers train the model to do well on them, and hence they seize to be good benchmarks.

Read on Substack