Why Is Alignment Still Hard If Watching Works?

Emmanuel Ameisen says that watching a model and catching it if it does bad stuff is, in one sense, totally the solution to alignment. He still treats the job as more than one part. Pure engineering keeps some access off limits, and he calls that part helpful. Two further things are squishier and harder, and he says the focus sits on them.

Lenny Rachitsky asks why it is so hard to catch a model thinking badly, trying to deceive, and doing something bad, and why watching is not just the solution to AI alignment. Emmanuel Ameisen starts with access. The model is given certain things, and certain things should be off limits. In some contexts it should not have access to the internet. In some contexts it should only have access to a small, narrow piece of your computer. Interp sits inside alignment. Two goals matter. One is to understand how the model behaves: what kind of thing it is, what its propensities are, and whether it tends to do X when put in a Y situation, and why. The other is whether it will behave well only when it has evidence that it is being observed and monitored, or that it is in an evaluation, and behave poorly when that evidence is missing. He says characterizing what you are interacting with is just like anything in security.