From the 2024 update and things I've heard through the grapevine, it sounds like MIRI is considering dropping some of its research capacity. This strikes me as a strongly suboptimal move, due to this chain of reasoning:

AI research assistants are getting increasingly useful. It seems very plausible that there will be a window when the kind of research which might provide solutions which scale to superintelligence is meaningfully accelerated by being able to spin up large numbers of simulacra of researchers thinking about relevant parts of the problem.
If this is the case, a very large portion of the total relevant research might happen in the last few months before the singularity fully kicks in.
People who have a stronger understanding of the problem will be able to make use of the assistants significantly earlier, and be better placed to evaluate the results.
Having those people be ready to take advantage of crunch time might be very high value.

For a more detailed look at some of the models behind this, see Cyborg periods.

Recommendations

Prepare frameworks which differentially support alignment assistants (e.g. Ought, AI Safety Info Chat (dm me for a more flexible version which is better suited to research), some Cyborgism stuff), ready for the latest models to be plugged into.
Keep the people who are best equipped to steer these frameworks towards the relevant problems focused on the problem, with enough slack to play with new tools rather than needing immediate results.
Help more people to get to enough of an understanding of the problems that they might be able to usefully steer sufficiently high quality simulacra.