I do AI Alignment research. Currently at METR, but previously at: Redwood Research, UC Berkeley, Good Judgment Project.
I’m also a part-time fund manager for the LTFF.
Obligatory research billboard website: https://chanlawrence.me/
LawrenceC
Karma: 6,558
Communicating with people who disagree on “obvious” things
An Angry Review of Greg Egan’s “Didicosm”
A “Lay” Introduction to “On the Complexity of Neural Computation in Superposition”
Informal Leadership Structures and AI Safety
Quality Matters Most When Stakes are Highest
Summarizing and Reviewing my earliest ML research paper, 7 years later
Post-mortem’ing my earliest ML research paper, 7 years later
Why I’m Less of a Shill for Related Work Sections
Two Examples of Joy in the Seemingly Mundane
For with what judgment we shall be judged
A Retrospective of Richard Ngo’s 2022 List of Conceptual Alignment Projects
5 Hypotheses for Why Models Fail on Long Tasks
Thanks for posting this—I was going to link Nikola’s work!
On not being scared of math
Have we already lost? Part 3: Reasons for Optimism
Have we already lost? Part 2: Reasons for Doom
Good catch, will fix.
Thanks for writing this!
As StanislavKyrm has written elsewhere, we have indeed tested Claude Code and Codex, and previously, we’ve done the same with other agent scaffolds in the literature.
Onto the other two problems you raise:
I don’t know if I’d say “egregious”. I agree the confidence intervals are large, but since they’re largely a result of uncertainties in the task suite, the differences between models are generally significant. Probably the real fix is a better way to present this uncertainty.
Yep, real problem. Opus 4.6′s “real” 50% time horizon is accurately described as “beyond the ability of METR to accurately measure with the Time Horizon 1.1 Suite”.
We’re working on it!
For example, we made a suite of tasks in collaboration with Epoch (MirrorCode), that unfortunately turned out to be 1) less representative of the tasks currently in the suite and 2) seems to have been partially saturated while it was being developed.
As I’ve written elsewhere, it’s pretty hard to make long tasks, and it’s not just a monetary expense.