kipply's blog

Old SSC again, on the virtue of silence.

the georgist roots of american libertarianism. descent of species. julia (this one is so grand). the garden of forking paths. secular shrine theory. maybe one of these days i will write up an essay about shinto-judaism but i’m hoping to try practicing it…

This Rob Stephenson almost cured my trauma about living in Alphabet City.

I like Thinking Machines, but more than that, I’m grateful.

Reread this iconic CS Lewis post called the Inner Ring.

Exciting Alignment Auditing paper from Anthropic!

There’s a lot of good insight in the Liang Wengfeng interview, most exciting to me that they don’t use OKRs!

Can’t believe I’ve never read the Scott Alexander’s Half An Hour Before Dawn In San Francisco it’s so good.

A checklist format of what we need to succeed at AI Safety, and it’s specifically safety and not alignment as there’s a pretty good case for a Control agenda.

i’ve been in a work hole these two months thinking about kernels and numerics. vitalik on techno-optimism, he’s amazing and this is well reasoned and written but also i am mildly annoyed that everyone stopped to look at it when nothing in it is that novel.

MMLU scores are getting too high — I’m optimistic about GPQA.

i think morality as a human trait is much shallower than people think, whether it’s about how nice you are to your friends, how much money you donate or your dietary choices. it’s much closer to how well people exercise (physically) as a trait, than say o…

An impressively redeeming interview with Mustafa Suleyman.

A Liang Lab evals of long contexts, for which we find that language models have the same biases humans do to the beginning and end.

A unique yet simple case for rogue AGI from Bengio.

Bleak month, busy month.

This document is my running literature review for people trying to catch up on AI.

Ben Kuhn on staring into the abyss as a core life skill.

Class Struggle is a board game designed by an NYU professor, Bertell Ollman.

Jan Leike characterizes possible solutions to the alignment problem, the most concerning aspect of which is eliciting human values imo.