Top postsTop post
Epistemic Status: I think each part of this piece is worth engaging with in its own right. I haven't thought as much about whether the ideas presented here add up to "it's more likely than not we live in a simulation". > There are more things in heaven and earth,...
Recent AI models are really reward hack-y. This is bad. It's the primary way in which current models are misaligned/dangerous/uncontrollable. The hypothesis put forth in this article is that they're like this because we're RL-ing them too hard. We're applying so much optimization pressure on programming and other capabilities that...
There are a number of reasons to believe current AI models are conscious. I mean “conscious” is the sense of “is there something it is like to be an AI model?” and “does the AI model have phenomenal experience?”. As to what “AI models” refers to, the short answer is...
The time it takes an AI or a Human+AI team (a "cyborg") to complete a task is a key aspect of what we care about when we talk about capabilities. The relationship between AI capabilities and cyborg capabilities is very useful for forecasting AI timelines. We simply don’t have good...
The low-background steel problem Modern steel is slightly radioactive. We did a lot of atomic testing in the 40s and 50s, and now our atmosphere has some amount of radioactive particles, which make their way into steel during production. This is mostly fine, but some scientific instruments require steel that...
[EDIT: LawrenceC, who works at METR, responds to this.] In this post, we’ll discuss three major problems with the METR eval and propose some solutions. Problem 1: The METR eval produces results with egregious confidence intervals, and the METR chart misleadingly hides this. Problem 2: There's a lack of sample...