All Posts

A breakdown of the GeoReasoner paper, which leverages both linguistic and geospatial information to reason on geospatially grounded natural language.

How I extracted 67 million words of legal text from 10B tokens of web data using heuristics and classifiers.

We test whether GPT-4 can translate agents’ simulated lives into readable, empathetic narratives—and show that style transfer beats ‘please be empathetic’ prompts.

Our new study asks if LLMs can learn enough conceptual modeling to pass graduate quizzes by using the same course materials as students.