I think that working on technical AI alignment is the most impactful thing I can do for the next five years of my life, for longtermist reasons. Finishing my MPhil at Cambridge. Research papers I'm working on: (1) meta-learning to train small language models on multilingual named entity recognition, (2) using transfer entropy to quantify collective behavior in feral horse harem herds, (3) measuring accuracy for question answering from large language internals, (4) detecting a circuit for subject-verb agreement in GPT-2, (5) teaching transformers how to do modular arithmetic, and (6) using neural networks as a measure of complexity for restricted minimum sorting problems.