A Stanford team injects two-hour qualitative interviews of 1,052 real individuals into an LLM and predicts their attitudes and behaviors at the individual level.

A Markovian classification of DSGE models that puts Partially Observable MDPs at the generalization of linear rational expectations, and derives an equilibrium as a fixed point mapping observables to beliefs.

A close reading of the 2023 UIST paper that fused memory streams, reflection, and recursive planning into an LLM loop, with the numbers and failure modes kept intact.