All of billzito's Comments + Replies

I don't think data companies can deliver on this complex of a task without significant oversight.

You may have already seen this, but in case you hadn't already, a related CFAR technique is internal double crux:

Strong +1 for writing the post-mortem -- there are incentives not to write that kind of thing, but I appreciated it. I also get value out of your weekly posts, so thanks for writing those.

That being said, I do have a few minor quibbles with the post-mortem. I think I interpreted your December (and future month) claims more strongly than it sounds like you interpreted them.

I haven't re-read all your posts carefully to check for places you might have said this, but I don't remember seeing you say in the 1-2 months after your "We're F***ed" post that you thou... (read more)

I agree that 55% would be an ambiguous evaluation, but 40% is substantially different than 65% and I've been using 40% in my recent models. Central mistake is the same in both cases. I don't remember what my update timeline was, I'd have to go back and look. Makes sense I would be too slow.

My personal experience in Uber/Lyfts (pre-covid, I used them multiple times a week for several years) is that they're probably more dangerous than driving myself (>80% of rides are very safe / normal, but the 1-20% where I think they're driving too fast or recklessly seem like most of the risk). I personally would be happy to pay 10% more to guarantee a safer driver, especially on e.g. a rainy day. I think I probably have more experience driving than most Berkeley EA's though / feel more confident in my driving skills.

"On the margin, if you don’t book an appointment, either the appointment and shot you decline will go unfilled, or it will probably go to someone else who is ‘high risk’ according to some list but unlikely to be actually high risk, or someone who is lying. In many jurisdictions all you have to do is say you are somehow eligible. That’s it. No one is verifying anyone’s claims.”

At least in the Bay Area, a couple hours of looking into this / thinking about it suggests this is false. I'm interested in counterarguments, but I think the pro-social thing to do is... (read more)

Seems like we should consider the possibility that the UK strain is not as transmissible as the pre-print suggested given the large drop in cases? Unless we have evidence that their lockdown is actually much more severe than similar lockdowns? A quick look at the lockdown rules suggests it’s similar to the Bay Area’s for instance.

[To clarify: I think this was worded somewhat incorrectly -- I didn't update much on the probability of the pre-print being wrong, but I did update significantly on the implications of the higher transmissibility.]

I’m in the UK. Rules are stricter than ever but also people are taking it seriously, more than the 2nd lockdown. And it’s January and freezing cold so no one wants to go out anyway.

Was your prediction that cases would go down and then up also for the UK in addition to the US? I thought the new strain prevalence was already pretty high in parts of England according to the pre-print, so it seems like cases going down by a lot there is a big update towards the pre-print potentially being wrong? The UK's last wave doesn't look naively out of distribution compared to e.g. Spain and the U.S.

[To clarify: I think this was worded somewhat incorrectly -- I didn't update much on the probability of the pre-print being wrong, but I did update significantly on the implications of the higher transmissibility.]

No, my numerical predictions are USA-only unless it says explicitly otherwise.

For "what task should I do next," it's O(n), because you just go down the list once doing a pairwise comparison. "What task should I do next" seems more important for actually doing things than sorting the entire list at first to avoid deliberation time / indecision / harder 3-way value comparisons.

Right, for a single pass it's a find-the-maximum-element algorithm in O(n). I think if you eventually do every task on the list it's equivalent to sorting the list? But this basically never happens to me.  Presumably intermediate states (doing e.g. half the items) is of intermediate efficiency? But my grasp of the underlying theory here is pretty weak.

Ya I think people should default to assuming it's out of date, but I do plan to check it semi-regularly and edit the post to make it clear when I last looked / if things have changed. Verified the routes still made sense today (8/25) and made it clear that was when I last checked.

I'm curious for how long this kind of list stays current. I could imagine things having shifted after several days or a week. Just a thought for people to check before following a route here.
  • How should we be thinking about second waves from May - Dec? In particular, what evidence should we look for that countries / states are well-prepared to handle potential outbreaks (e.g. even countries like Singapore are having problems with second waves right now)?
  • +1 “What is my risk from delivery food, given a certain prevalence?" And the same question for packages and delivered produce, given cleaning them or not. I thought about this for ~1hr but didn't generate a great answer, I think it'd save a bunch of people time to have a good one.
  • What are the chances that various vaccine efforts will work in x months (in particular, chances of working in <12 months)?

I can't speak for habryka, but I think your post did a great job of laying out the need for "say oops" in detail. I read the Doomsday Machine and felt this point very strongly while reading it, but this was a great reminder to me of its importance. I think "say oops" is one of the most important skills for actually working on the right thing, and that in my opinion, very few people have this skill even within the rationality community.

Glad that you found something that worked for you! I used to struggle with trying to control my sweet tooth, but found that going "no sweets" was surprisingly effective as I never had to decide. The decision is where I've had problems in the past because if I'm excited/stressed/etc., I decide to have too many sweets.

I've experimented with different versions of "no sweets" over the last three years, including no sweets except one cheat day a year, no sweets except one cheat day a month, no sweets except one-three bites of ... (read more)

Thats fabulous! I have been taking very tentative nibbles of people's desserts, and I'm glad to hear your 1-3 bites strategy is working for you because that's the strategy that I'd most like to use as well! (Social food is so important!)