Compute verification might be a very significant part of coordination with respect to intelligent AI systems. In short timelines (<2030), the kind espoused by a frontier lab leader, I'm suspicious of die-level verification hardware being tractable. If this is the case, we should then focus on auxilary verification methods, such...
Recently, AI Models have become good at long horizon tasks. In addition, they seem especially good at the types of long horizon tasks that allow for a quick and short feedback loop. For example, on MirrorCode, models were able to generate tens of thousands of lines of code, a task...
I think that preventing suffering is more important than causing happiness, and I try my best to prevent the suffering of all things that I consider moral subjects. To this extent, I'm vegan, donate my money to effective charities, and so forth. At the time of writing, I'm thinking a...
In this post, I claim a few things and offer some evidence for these claims. Among these things are: * Language models have many redundant attention heads for a given task * In context learning works through addition of features, which are learnt through Bayesian updates * The model likely...