Key Info

Martin Casado argues that labs missed Jev because they built chat-based LLMs that generate text, while software needed a model that chooses among options, which Jev does incredibly fast, cheaply, and accurately.

Highlights

  • Casado criticizes the jankiness of trying to fit text-generating LLMs into traditional software.
  • Jev's approach: given a set of options, it selects the best one, trained specifically for that task.
  • Claims this has seen the fastest adoption, implying a shift from text generation to action selection.