This is a response to Gwern's Guardian Angels post and draws terms from it quite extensively.Epistemic status: draft and to-be-updated once I do more experiments.Suppose you…
If Bayesian stats is so great, it should be able to do everything that ordinary stats can do. Indeed, as a fan of Bayesian stats, I…
In AI control[1] research, we've developed many protocols: trusted monitoring, untrusted monitoring, resampling, etc. They each have a different safety-usefulness tradeoff, and labs might use a…
When I was in middle school, one of our teachers gave us a “don’t do drugs” talk.Somebody asked him whether he had ever used drugs himself.…
We have a new standard policy for releasing frontier AI models. It is not good. We are now, it seems, going to have the White House…
Alternative titles:… earns you a high-risk pregnancy… earns you an ascent of Matterhorn… earns you 10,000 km on a motorcycle… earns you two BASE jumps… earns…
The existential AI safety community needs to take building a civic and social movement seriously as a core intervention. We believe this is a high-value, badly…
Car crashes kill over 35,000 people in the US every year. Plane crashes, on the other hand, kill ~350. Despite this, we have shows like Mayday/Air…
This work was done as part of the BlueDot's Technical AI Safety Project Sprint and should be treated as an informal report of preliminary results done…
In AI catastrophe, KatjaGrace writes about how difficult it can be for some people to take X-risk seriously:My guess is that an important thing going on…