Conditioning as a Crux Finding Device
Say you disagree with someone, e.g. they have low pdoom and you have high pdoom. You might be interested in finding cruxes with them.
You can keep imagining narrower and narrower scenarios in which your beliefs still diverge. Then you can back out properties of the final scenario to identify cruxes.
For example, you start by conditioning on AGI being achieved - both of your pdooms tick up a bit. Then you also condition on that AGI being misaligned, and again your pdooms increase a bit (if the beliefs move in opposite directions, that might be worth exploring!). Then you condition on the AGI self-exfiltrating, and your pdooms nudge up again.
Now you've found a very narrow scenario in which you still disagree! You think it's obvious that a misaligned AGI proliferating around the world is an endgame, they don't see what the big deal is. From there, you're in a good position to find cruxes.
(Note that you're not necessarily finding the condition of maximum disagreement, you're just trying to get information about where you disagree.)
Got it thanks!
(eg. any o1 session which finally stumbles into the right answer can be refined to drop the dead ends and produce a clean transcript to train a more refined intuition)
Do we have evidence that this is what's going on? My understanding is that distilling from CoT is very sensitive—reordering the reasoning, or even pulling out the successful reasoning, causes the student to be unable to learn from it.
I agree o1 creates training data, but that might just be high quality pre-training data for GPT-5.
Why does it make the CoT less faithful?
Favorite post of the year so far!
My favored version of this project would involve >50% of the work going into the econ literature and models on investor incentives, with attention to
- Principal-agent problems
- Information asymmetry
- Risk preferences
- Time discounting
And then a smaller fraction of the work would involve looking into AI labs, specifically. I'm curious if this matches your intentions for the project or whether you think there are important lessons about the labs that will not be found in the existing econ literature.
1How does the fiduciary duty of companies to investors work?
OpenAI instructs investors to view their investments "in the spirit of a donation," which might be relevant for this question.
I would really like to see a post from someone in AI policy on "Grading Possible Comprehensive AI Legislation." The post would lay out what kind of safety stipulations would earn a bill an "A-" vs a "B+", for example.
I'm imagining a situation where, in the next couple years, a big omnibus AI bill gets passed that contains some safety-relevant components. I don't want to be left wondering "did the safety lobby get everything it asked for, or did it get shafted?" and trying to construct an answer ex-post.
MIT Tech Review doesn't break much news. Try Techmeme.
Re "what people are talking about"
Sure, the news is biased toward topics people already think are important because you need readers to click etc etc. But you are people, so you might also think that at least some of those topics are important. Even if the overall news is mostly uncorrelated with your interests, you can filter aggressively.
Re "what they're saying about it"
I think you have in mind articles that are mostly commentary, analysis, opinion. News in the sense I mean it here tells you about some event, action, deal, trend, etc that wasn't previously public. News articles might also tell you what some experts are saying about it, but my recommendation is just to get the object-level scoop from the headline and move on.
Re is it worth the time of sifting through
Skimming headlines is fast. Maybe the news tends to be less action-relevant for your research, but I bet AI safety collectively wastes time and misses out on establishing expertise by being behind the news. Reading Zvi's newsletter falls under what I'm advocating for (even though it's mostly that what-people-are-saying commentary, the object-level news still comes through.)