Google AI PM; Foundation board member
Dave Orr
Karma: 501
Together for 21 years, married for 17, two kids, all good.
The major thing I had to learn was how to communicate certain things, and when to keep my mouth shut. For example, she doesn’t ascribe to the principle of charity when it comes to strangers or the out group. We used to have fights about it (object level always, “why are you taking their side?!?”) before I realized they were fundamentally unproductive. And I can think of a couple of key times that she was right and I was overly charitable, including in a work context that really mattered.
On the EA front, we donate 10% of income, to a mix of things—hers more socially determined, mine more EA-ish, but good organizations all.
I think the world is full of people who don’t t think the way you do, no matter who you are, so it’s important to be able to form relationships with lots of kinds of people. Hopefully some experience/interest in rationality could help identify otherwise puzzling communication failures.
Also, practically speaking, less wrong / EA / rationalism are heavily male, so most people will need to find someone outside the community as a life partner.
I’ll chime in with my experience, presumably from the same employer.
I and my team at work have found a pretty high rate of failures, in one case a majority of tests failed. Seems like the failure rate is around 30%, sample size of ~8 people, though not sure of how many tests/person. So that $60/test might look more expensive.
Having said that, I really like the Cue tests. More accurate than fast tests, and super convenient. I’m not very price sensitive so for me it’s a clear win.
This is the Godel Escher Bach solution :)
This was my solution! :)
I found a different solution to the initial puzzle, which I won’t spoil here, but post as a follow-on:
Same scenario, except after you hear the statement, you know the person is single—but you don’t know who they worship!
What was the statement?
https://astralcodexten.substack.com/
FKA slatestarcodex.
Of course tons of this research is going on. Do you think people who work at Facebook or YouTube are happy that their algorithms suggest outrageous or misleading content? I know a bit about the work at YouTube (I work on an unrelated applied AI team at Google) and they are altering metrics, penalizing certain kinds of content, looking for user journeys that appear to have undesirable outcomes and figuring out what to do about them, and so on.
I’m also friends with a political science professor who consults with Facebook on similar kinds of issues, basically applying mechanism design to think about how people will act given different kinds of things in their feeds.
Also you can think about spam or abuse with AI systems, which have similar patterns. If someone figures out how to trick the quality rating system for ads into thinking this is a high quality ad, then they’ll get a discount (this is how e.g. Google search ads works). All kinds of tricky things happen, with web pages showing one thing to the ad system and a different thing to the user, for instance, or selling one thing to harvest emails to spam for something else.
In general the observation from working in the field is that if you have a simple metric, people will figure out how to game it. So you need to build in a lot of safeguards, and you need to evolve all the time as the spammers/abusers evolve. There’s no end point, no place where you think you’re done, just an ever changing competition.
I’m not sure this provides much comfort for the AGI alignment folks....
I think seasonality is going to push in the other direction for a while. I think in 7-8 months things could plausibly be much better, in fact I think it’s likely, but February is still very winter.
Out of curiosity, when do you crosspost to LW versus just posting to ACX? I can see that this post is very LW since it’s about biases and thus rationality, but I have to think almost all the readers of LW also read ACX.
Is the discussion in the comments different here? (Certainly the comment interface and threading are way better here.)
Off the cuff: In Alameda County, I’d maybe put 2⁄3 probability on at least 85% of weeks being between 5000 official new cases per day and 80,000 official new cases per day, until November?
That’s a good way to think about it. My 2/3rds confident window looks something like 1.5k-25k per day for most weeks until November. So substantially lower than yours.
For example, if we had the same infection rate as the UK (cases are down from delta peak) it would be 500 cases/day in Alameda county. The UK might be trending back up now, but the slope is much lower. Their vaccination rate is a little higher than Alameda county but our vaccines are better, so that’s probably ~ a wash, or maybe slightly in the UK’s favor.
this Dec 15 - Jan 15 average will be at least 50,000 cases per day
In other words, you think the average for that month will be equivalent to the worst week last winter before anyone was vaccinated? Definitely seems high to me. Delta is much worse, yes, but vaccines really do help quite a bit, I think.
I buy that seasonality is a thing for covid, but I think it will be more like 15kish rather than 50k for that month.
Based on this, I expect this coming Nov/Dec/Jan/Feb to be way worse again around SF. Except the baseline is much higher now, so I expect the elevated case rate to likewise be much higher
I’d like to register a prediction that in the Bay Area and in CA, rates will be drastically below where they are today in a couple of months.
The reason I think that is that everywhere else, we’ve seen delta skyrocket, and then collapse once the control system kicks in and herd immunity thresholds are reached as vulnerable people get infected. I think the same thing will happen here.
So contra the advice that this is the best time to do things, I think it’s worth waiting. If I’m right we should see evidence in the next few weeks.
The long run stock market returns are just below 8% (https://www.wealthsimple.com/en-ca/learn/average-stock-market-return#:~:text=Longer-term returns,-2018 is about 7.96%.), so I think your 12% is optimistic. Your “low returns” case is I think more accurate.
I’m not sure anything in your post really depends on the numbers, but that extra 4 points is a lot of money in the out years.
For a fascinating, borderline nsfw look at bear week during COVID, and why infections might be atypical, this reddit thread is worth a gander.
“My point is… To everyone worried about the P-Town data: I wouldn’t get too nervous going to the grocery store just yet—unless you tend to have orgies at Market Basket.”
I think characters via alphabets dwarfs everything else for written language, but here are a few other factors:
how phonetic it is if you want to learn the writing system.
how regular it is—English is full of exceptions (because of its 3 language family history). Spanish is very regular, Russian also has a ton of exceptions.
how complicated the morphology is (Finnish is tough because of the super complex morphology, see also Russian). Mandarin is very easy on this dimension.
whether there’s a phoneme distinction that you didn’t learn as a child—so for a Japanese speaker, l vs r is hard in English (“Engrish”), and for an English speaker, o versus ō is hard in Japanese.
In general, of course, the more similar it is to your native tongue, the easier it is. Tones are hard to learn if you don’t speak a tonal language, but if you do then they are super intuitive. Similar with lots of morphology, fixed versus fluid word orderings, etc.
The other angle is spoken versus heard. Portuguese (especially) and French are much easier to speak than understand because of the various ways that sounds are elided or mushed together with fluent speakers. So you can get basic sentences out before you can understand something at full speed—generally true but much more so for some languages.
I think that’s a fair criticism.
I work on NLP AI systems and have spent a lot of the past decade working on developing training data, so I have a degree of expertise here.
There are a lot of things that go into how hard a language is to learn. How close the spelling is to pronunciation is one of them, but not the dominant one for alphabetic languages (grammatical gender is another, which you don’t mention).
But you are totally correct that learning a character-based system is much, much harder than learning an alphabetic (or syllabary) language. I think it’s a huge disadvantage for e.g. China, where kids have to devote years of study and memorization to be literate, many more than in alphabetic languages. I would go further and say that they are forced to have their whole school system revolve around immense amounts of rote memorization, which I think could lead to the kinds of relative lack of creativity that the Chinese school systems are sometimes criticized for. I don’t think you can learn written Chinese without massive amounts of memorization, leaving little room for other ways of thinking for a decade or more.
Further reading:
Difficult and easy languages (Chinese is easy to learn to speak, super hard to learn to read/write)
Of languages that got multiple votes, Catalan and Spanish (and Esperanto) rated as very easy, and written Chinese (and especially literary Sinitic which is old written Chinese) rated very hard.
I guess I’m just dense here, but I still don’t see how it can be that the risk adjusted return on capital is unaffected by taxes. Borrowing money (i.e. leverage) adds risk so that can’t be it (or there’s an additional mechanism that comes into play). Later you say that the government is a partner but they aren’t reducing your risk, they’re just taking half your profits.
Probably not worth the back-and-forth more here but to me the “taxes don’t affect returns” position is just obviously wrong and nothing you’ve said shows a mechanism that would change that.
Investors are not profit-maximizing. Investors are (arguably) risk-adjusted profit-maximizing.
Is there enough money in the world for all investments to lever up 100%? There’s certainly not enough that the borrowing costs would be trivial, if debt demand were suddenly so high.
Also, 100% leverage doubles the risk for the same return (by hypothesis) which probably needs some more support before it’s clear that that is socially better compared to status quo. Note that many investment strategies get totally wiped out (due to gambler’s ruin) if risk gets too high for the same return.
A better model is that investment capital seeks the best risk adjusted return. Right now there’s a balance between opportunities in debt, equities, real assets, etc. If you increase taxes and therefore decrease return on equities, enough capital will move out to other asset classes until the risk adjusted returns are roughly equal.
Maybe that new equilibrium is better or maybe it’s worse, but denying that it will change I think makes your analysis hard to accept.
Thanks for all the work you put in on these incredibly informative posts. This is, hands down, the best source of analysis I’ve found for all things COVID.