See also Anthropic's view on this 

[seeing] a lot of safety research as "eating marginal probability" of things going well, progressively addressing harder and harder safety scenarios.

The implicit strat (which Olah may not endorse) is to try to solve easy bits, then move on to harder bits, then note the rate you are progressing at and get a sense of how hard things are that way. 

This would be fine if we could be sure we actually were solving the problems, and also not fooling ourselves about the current difficulty level, and if the relevant research landscape is smooth and not blockable by a single missing piece. 

Nitpick: It's fairly unlikely that GPT-4 is 1tn params; this size doesn't seem compute-optimal. I grant you the Semafor assertion is some evidence, but I'm putting more weight on compute arithmetic.

Vouching for this, it's a wonderful place to work and also to hang out.

A successor project is live here, takes all comers.

Scottish degrees let you pick 3 very different subjects in first year and drop 1 or 2 in second year. This seems better to me than American forced generalism and English narrowness.

Thanks: you can apply here.

I've edited the post to link to the successor project.

I dream of getting a couple questions added onto a big conference's attendee application form. But probably not possible unless you're incredibly well-connected.

Oh that is annoying, thanks for pointing it out. I've just tried to use the new column width feature to fix it, but no luck.

Here's a slightly more readable gdoc.

it is good to omit doing what might perhaps bring some profit to the living, when we have in view the accomplishment of other ends that will be of much greater advantage to posterity.

- Descartes (1637)

Yes, if I was using the same implicature each time I should have said "MacAskill" for Guzey. Being associated with Thiel in any way is a scandal to some people, even though his far-right turn was after this talk.

It's not normative, it's descriptive - "shameable", not "ought to be ashamed".