The classic "Dallas" example from Anthropic focuses on an internal circuit in an LLM.I became curious about what the same underlying process looks like when viewed…
Spoiler: positive expected utility. That’s what it’s good for.In The War Trap[1], Bruce Bueno de Mesquita lays out a theory of war that beats all other…
Our kids like listening to music on their tablets, and we decided it's worth it to pay some money so they can make playlists and don't…
I'm wondering what comes to their mind when they hear 'AI-Risks' or 'AI-Safety'. At this time, people had been using AI, and then they realized that…
SummaryWe train a new token—a neologism (Hewitt et al.)—for a model, but unlike Hewitt et al., we train it on data the model generated while steered…
I often vocally argue for the degendering of English Pronouns, that is to say, the replacement of the pronouns he/she with ungendered singular equivilents (depending on…
[This is an introductory blog for the paper Laguerre Geometry for Interpreting Large Language Models and the GitHub repository Geometric Lens.]LLM Lens: What does an internal…
The contest deadline was today. Here's what I submitted. Original here.Contest submission: Epistack-HowTruthfulThis is a submission to FLF's Epistemic Case Study Competition.If you're not a contest…
tl;dr - Steering vectors, inoculation prompting, and post-hoc honesty fine-tuning can all be understood as variants of one alignment strategy, which I call train-deploy mismatch. Each…
Many of the content creator fellows at plzdontkillus found my thoughts useful when I visited two weeks ago, so I’m now sharing a write-up here.Many thanks…