The case for satiating cheaply-satisfied AI preferences
A central AI safety concern is that AIs will develop unintended preferences and undermine human…
Category: Opinion
Immortality: A Beginner’s Guide (Part 2)
This is the second post in my chain of reflections on immortality, where I will…
Investigating encoded reasoning in LLMs
Epistemic status: This work was done as a 1-week capstone project for ARENA. It highlights…
Claude Code, Claude Cowork and Codex #5
It feels good to get back to some of the fun stuff. The comments here…
Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation
TL;DR: We introduce a testbed based on censored Chinese LLMs, which serve as natural objects…
Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation
TL;DR: We introduce a testbed based on censored Chinese LLMs, which serve as natural objects…
Emergent Misalignment and the Anthropic Dispute
TL;DR: We think allowing frontier AI models to be used for mass domestic surveillance and…
Might An LLM Be Conscious?
Might An LLM Be Conscious?There’s no scientific consensus on whether current or future AI systems…
Mapping AI Capabilities to Human Expertise on the Rosetta Stone (Epoch Capabilities Index)
This is a crosspost from the General-Purpose AI Policy Lab research blog.The “Rosetta Stone for…
Intro: Non-Identifiability of Explanations
This is the introduction to the Which Circuit is it? sequence. We will develop some…