maxc.life

Research

Small, independent studies in two strands: how language models handle questions that don't have one checkable answer (is this ad misleading, does this rule fit these facts), and what New Zealand's statute book shows when you read all of it. Each one sets out the question, the method, what it found and where it falls short.

The inspiration is Anthropic's Conceptual Reasoning Fellowship, where experts in law, policy, economics and philosophy turn judgement-heavy questions from their field into tests of how well a model reasons. These projects borrow the shape of that work at a much smaller scale: one narrow question at a time, built in my own time, with real decisions as the benchmark.

Misleading Ads Eval

Research · 2026 · In progress

Can a language model tell when an ad is misleading the way New Zealand's Advertising Standards Authority would? A small, open evaluation built on published Complaints Board decisions.

Who makes New Zealand's law?

Research · 2026 · In progress

Parliament passes the Acts, but more and more of the detail is made by Ministers and agencies. Is law-making moving away from Parliament, and is Parliament's scrutiny keeping up? A study built on the full statute book.