Map and Territory, Predictably Wrong (2)
Crossposted (with small tweaks) from my Substack.Burdensome DetailsThis post explains the conjunction fallacy, i.e., people sometimes assign a higher probability to A and B together than…
Latest
LessWrong App
Hello everyone long time lurker here. I know most people here probably prefer a PWA but if you are like me and prefer an app, I…
Chill Buddhism
Inspired by reading some of the comments and discussion around some of the recent meditation controversies but also a general reflection on the way that people…
Questions for the reader (and writer) of "Fundamental Uncertainty"
This is a commentary relating to Gordon Worley's book "Fundamental Uncertainty". There was a contest with prizes, and there are at least three other entries available…
Agnostic Meaning: Betting on Others
Epistemic Status: The decision process here is intentionally vague, and is designed to be a short-term launchpad for those new to EA, or those going through…
The Art of Shipping Slopware
Meta famously created an internal AI-usage leaderboard in pursuit of tokenmaxxing. I thought this backwards incentive structure was an anomaly until my friend who works at…
A Chess Training App for Imitation Learning
LLMs learn faster if we first pretrain them to imitate dense teacher-forced examples. I speculated that this would work on humans too, so I built a…
MUD as AI Evaluation and LLM-judge distortion in ways aggregate κ misses
A group of friends and I spent the last several months running an experiment in our free time to determine if a MUD would be a…
Constitutional Midtraining: Content Presence Drives Alignment Gains
A more accessible, much shorter version of our paper that goes by the above title. Paper here. Code and benchmarks here. Data and models here. Would…
RLVR that rewards red teaming the training environment
Epistemic status: seeing what sticksI've been thinking pretty obsessively about how to make sure the Hugging Face incident doesn't happen again. I don't work at a…