AS ROCKS MAY THINK

Summary from the TLDR newsletter

LLM-based reasoning is both powerful and flexible. However, there is likely still room to make the recipe even simpler. There are also still more gains to be had by understanding all the plays in which thinking behavior can be found. The extra reasoning data generated over recent years may help further train models to reason better through previously unviable methods.