Featured Articles

Survey of prescient actions

In a 10-20 hour exploration, we did not find clear examples of ‘prescient actions’—specific efforts to address severe and complex problems decades ahead of time and in the absence of broader scientific concern, experience with


Takeaways from safety by default interviews

Last year, several researchers at AI Impacts (primarily Robert Long and I) interviewed prominent researchers inside and outside of the AI safety field who are relatively optimistic about advanced AI being developed safely. These interviews


Atari early

Deepmind announced that their Agent57 beats the ‘human baseline’ at all 57 Atari games usually used as a benchmark. I think this is probably enough to resolve one of the predictions we had respondents make


Three kinds of competitiveness

[Epistemic status: I wrote this for Blog Post Day II. Sorry it’s late.] In this post, I distinguish between three different kinds of competitiveness — Performance, Cost, and Date — and explain why I think


AGI in a vulnerable world

I’ve been thinking about a class of AI-takeoff scenarios where a very large number of people can build dangerous, unsafe AGI before anyone can build safe AGI. This seems particularly likely if: It is considerably

AI Timelines

2019 trends in GPU price per FLOPS

We estimate that in recent years, GPU prices have fallen at rates that would yield an order of magnitude over roughly: 17 years for single-precision FLOPS 10 years for half-precision FLOPS 5 years for half-precision

No Picture
Continuity of progress

Incomplete case studies of discontinuous progress

This is a list of potential cases of discontinuous technological progress that we have partially investigated. List In the course of investigating cases of potentially discontinuous technological progress, we have collected around fifty suggested instances