← Index

CHOOSEOther

LLM Travel Demand Model Evaluation

Sumit B@relentlessSB
Project photo 1

This project evaluates if an LLM-based decision model, TypeSafe's Jev, can substitute for a traditional travel demand model by comparing its outputs to Census commute data across US metros.

I tested whether an LLM-based decision model (TypeSafe's Jev) can replace a travel demand model. 211 US metros, name and population only, vs Census commute shares. Ask "how does a typical worker commute?" and it says "drives alone" 99% everywhere (blue). Ask "what share of workers drive / take transit / walk?" and it tracks the Census, r = 0.8 to 0.9 (red). Same model. Levels yes, individual choices no. @typesafeai
Original post