Legal informatics and AI
Background
I came to law school after spending several years doing data analysis and modeling in the physical sciences and finance. In contrast to those fields, legal practice seemed old-fashioned. How could I automate a process? How could I make sense of everything I was reading? Are we still in the nineteenth century or what? For all of its zeal to tackle new problems, law remained in many ways stuck in the past when it came to actually doing the work. It was frustrating, and I even wrote a Note about it as a kind of protest.
We’re now in a different era, and the current crop of transformer-based language models promises to make lots of ideas that were enticingly out of reach a decade or two ago much closer. How can we take advantage of the new AI tools?
AI, generally
- Bishop, C. and Bishop, H. (2024) Deep Learning: Foundations and Concepts. Springer. Authors’ site.
- Jurafsky, D. and Martin, C. (2025) Speech and Language Processing (3rd ed. draft Aug. 24, 2025). Full draft here.
- Russell, S. and Norvig, P. (2021) Artifical Intelligence: A Modern Approach (4th ed.). Pearson. Authors’ site.
Publications & conferences
Useful tech for LLM hacking
- DeepEval: open-source LLM evaluation framework.
- Hugging Face Transformers: swiss army knife for transformer models.
- Instructor: library for structured LLM outputs.
- Marimo: reactive Python notebooks with bells and whistles.
- LLM: Simon Willison’s CLI and library for interacting with LLMs.
- uv: a modern Python package manager.