Research Notes

Notes on language models, cognition, safety, and human-centered AI.

Domain-Specific Evaluations With Real Consequences

Benchmarks for finance, medicine, and cybersecurity that mirror the consequences that matter.

2 min read · February 20, 2026

2026 · evaluation finance cybersecurity visualization · research evaluation
Teaching Language Models to Grow Up

BabyLM and psych-inspired batteries that treat LLMs as computational subjects of cognitive science.

2 min read · February 20, 2026

2026 · cognition babylm developmental-alignment llm · research cognition
Beyond the Unlearning Mirage

Dynamic probes and activation analysis that expose brittle unlearning, paired with watermarking and continual learning.

2 min read · February 20, 2026

2026 · safety unlearning watermarking continual-learning · research safety
LLM Copilots for Peer Counselors

How motivational interviewing–aware sandboxes and analytics help peer counselors level up.

2 min read · February 20, 2026

2026 · mental-health counseling llm feedback · research mental-health