I’m still getting better about moving my half-formed writing from my personal notes to something somewhat presentable to the public. Here’s what I have so far.

• Mitigating inner misalignment
  What happens when gradient descent (an optimizer) produces a model that is itself an optimizer (i.e. a mesa-optimizer)? How can we ensure alignment between the base optimizer objective and the mesa-optimizer objective?
• Proving the absence of deceptive misalignment
  An answer I gave to an interview question about designing a model that avoids deceptive alignment.
• Absolute morality and AI freedom
  Are us humans satisfied with another agent successfully achieving its goals (assuming they don't interfere with us) if those goals are not human?
• Considerations for building AGI - from a newbie to the field
  This was written in 2023 when I had just started gaining exposure to ideas around AGI and alignment. I wrote this as part of an assignment for an engineering ethics class.