[WIP] Incentives in AI Safety
We are in a period of unprecedented growth in AI safety, but resources and media attention concentrate heavily on harms caused by accidental model failures a...
Posts by Year
2026
2025
Mentors, Advisors, Advocates (and Allies)
Your job is not your career
2024
Interpreting vision models with sparse dictionary learning: a case for hierarchical learning
Recently, Anthropic demonstrated the power of sparse dictionary learning as an interpretability tool in a large language model. They applied the method to a ...
Mitigating LLM Sycophancy
A sycophantic AI is easily manipulated by the user to give an incorrect response. The model provides the correct answer to a neutral prompt, but pri...
Demonstrating LLM Sycophancy
A sycophantic AI is easily manipulated by the user to give an incorrect response. The model provides the correct answer to a neutral prompt, but pri...
2023
Getting a job in a new field
Here is all my advice about getting a job, particularly if you are coming from academia or switching fields. If you are considering a career change, I recomm...
2021
Failing repeatedly
What we (don’t) say about failure