Do you do weekly or daily reviews? What are they like? 2019-08-05T01:23:43.351Z
benwr's unpolished thoughts 2019-07-29T02:18:14.366Z
Why I've started using NoScript 2019-05-15T21:32:20.415Z
Usernames in RSS feeds 2018-02-19T02:22:26.749Z


Comment by benwr on benwr's unpolished thoughts · 2021-06-10T21:16:52.133Z · LW · GW

A thing that feels especially good about this way of thinking about things is that it feels like the kind of problem with straightforward engineering / cryptography style solutions.

Comment by benwr on benwr's unpolished thoughts · 2021-06-10T20:19:53.748Z · LW · GW

I'm interested in concrete ways for humans to evaluate and verify complex facts about the world. I'm especially interested in a set of things that might be described as "bootstrapping trust".

For example:

Say I want to compute some expensive function f on an input x. I have access to a computer C that can compute f; it gives me a result r. But I don't fully trust C - it might be maliciously programmed to tell me a wrong answer. In some cases, I can require that C produce a proof that f(x) = r that I can easily check. In others, I can't. Which cases are which?

A partial answer to this question is "the complexity class NP". But in practice this isn't really satisfying. I have to make some assumptions about what tools are available that I do trust.

Maybe I trust simple mathematical facts (and I think I even trust that serious mathematics and theoretical computer science track truth really well). I also trust my own senses and memory, to a nontrivial extent. Reaching much beyond that is starting to feel iffy. For example, I might not (yet) have a computer of my own that I trust to help me with the verification. What kinds of proof can I accept with the limitations I've chosen? And how can I use those trustworthy proofs to bootstrap other trusted tools?

Other problems in this bucket include "How can we have trustworthy evidence - say videos - in a world with nearly perfect generative models?" and a bunch of subquestions of "Does debate scale as an AI alignment strategy?"

This class of questions feels like an interesting lens on some things that are relevant to some sorts of AI alignment work such as debate and interpretability. It's also obviously related to some parts of information security and cryptography.

"Bootstrapping trust" is basically just a restatement of the whole problem. It's not exactly that I think this is a good way to decide how to direct AI alignment effort; I just notice that it seems somehow like a "fresh" way of viewing things.

Comment by benwr on Prize: Interesting Examples of Evaluations · 2020-11-29T14:02:15.206Z · LW · GW

IT security auditing; e.g.

Comment by benwr on Prize: Interesting Examples of Evaluations · 2020-11-29T00:27:01.702Z · LW · GW

"Postmortem culture" from the Google SRE book:

This book has some other sections that are also about evaluation, but this chapter is possibly my favorite chapter from any corporate handbook.

Comment by benwr on Prize: Interesting Examples of Evaluations · 2020-11-28T23:05:55.917Z · LW · GW

Two that are focused on critique rather than evaluation per se:

Comment by benwr on benwr's unpolished thoughts · 2020-09-27T04:26:02.606Z · LW · GW

If I got to pick the moral of today's Petrov day incident, it would be something like "being trustworthy requires that you be more difficult to trick than it would be worth", and I think very few people reliably live up to this standard.

Comment by benwr on benwr's unpolished thoughts · 2020-09-22T05:45:19.692Z · LW · GW

Beth Barnes notices: Rationalists seem to use the word "actually" a lot more than the typical English speaker; it seems like the word "really" means basically the same thing.

We wrote a quick script, and the words "actually" and "really" occur about equally often on LessWrong, while Google Trends suggests that "really" is ~3x more common in search volume. SSC has ~2/3 as many "actually"s as "really"s.

What's up with this? Should we stop?

Comment by benwr on Did any US politician react appropriately to COVID-19 early on? · 2020-07-17T02:25:44.508Z · LW · GW

San Francisco's mayor, London Breed, declared a state of emergency in the city on February 25th, and it seems like she was concerned about the disease (and specifically ICU capacity) as early as January.

I don't know what actions the mayor's office actually took during this time, but it seems like she was at least aware and concerned well ahead of most other politicians.

Comment by benwr on benwr's unpolished thoughts · 2020-07-16T23:10:49.002Z · LW · GW

darn - I've been playing it on my old ipad for a long time

Comment by benwr on benwr's unpolished thoughts · 2020-07-15T00:36:07.229Z · LW · GW

Recently I tried to use Google to learn about the structure of the human nasal cavity & sinuses, and it seems to me that somehow medical illustrators haven't talked much to mechanical draftspeople. Just about every medical illustration I could find tried to use colors to indicate structure, and only gave a side-view (or occasionally a front view) of the region. In almost none of the illustrations was it clear which parts of your nasal cavity and sinuses are split down the middle of your head, vs joined together. I still feel pretty in-the-dark about it.

In drafting, you express 3d figures by drawing a set of multiple projections: Typically, you give a top view, a front view, and a side view (though other views, including cross-sections and arbitrary isometric perspective, may be useful or necessary). This lets you give enough detail that a (practiced) viewer can reconstruct a good mental model of the object, so that they can (for example) use their machine shop to produce the object out of raw material.

There's a pretty fun puzzle game that lets you practice this skill called ".projekt"; there are probably lots more.

Comment by benwr on Coronavirus: Justified Practical Advice Thread · 2020-03-06T21:16:09.363Z · LW · GW

It is possible to both rinse your phone and put copper tape on it.

Comment by benwr on Coronavirus: Justified Practical Advice Thread · 2020-03-04T00:21:16.389Z · LW · GW

Yeah, you're right that I imputed a particular mechanism that isn't supported by the Wikipedia page - thanks for pointing that out. I do still think that the ions-getting-on-things mechanism is part of the story, mostly because the reduction sizes are really large. This could indicate either (a) that most microbes end up on surfaces first via touch surfaces, and spread from there, or (b) that copper ends up on nearby surfaces. Or some of both.

In this particular case, though, I think it's quite likely (because I've seen my hands turn a bit blue) that in fact copper and copper oxides are getting on my hands as a result of the tape.

Comment by benwr on How to fly safely right now? · 2020-03-03T22:01:50.248Z · LW · GW

Oh, whoops, I somehow misread your original post ("shave" was edited to "have" in my brain)

Comment by benwr on How to fly safely right now? · 2020-03-03T21:19:02.642Z · LW · GW

Another random thought: wearing gloves is probably not a great idea in the default case, but if you wear gloves and then replace them, it seems like that's probably about as good as washing your hands or using hand sanitizer (and might be easier to do often when sitting in your seat on the plane; it might also dry out your hands less)

Comment by benwr on How to fly safely right now? · 2020-03-03T21:14:38.078Z · LW · GW

Another thing you might want to bring: swimming or safety goggles. Iiuc, it's possible for viruses to spread via your conjunctivae (although this is probably a much smaller effect than transmission via nose or mouth).

Comment by benwr on How to fly safely right now? · 2020-03-03T21:11:18.860Z · LW · GW

Re facial hair: it's recommended not to have facial hair when wearing a face mask, because it makes the seal weaker. Do you think the protective effect is larger than that effect?

Comment by benwr on How to fly safely right now? · 2020-03-03T21:09:10.301Z · LW · GW

You can buy bottles for arbitrary gels, and then put hand sanitizer in those bottles.

Comment by benwr on Coronavirus: Justified Practical Advice Thread · 2020-03-01T23:01:19.290Z · LW · GW

A thing I probably haven't thought enough about is, "how much this will impact your rate of copper ingestion, and is that very bad?" My guess is that this is less important than the effects on infectious disease; it seems like it would need to increase your copper consumption by 100x in order to produce major negative health effects ( I may try to be virtuous and do a fermi on this later but also I'd welcome someone else trying to do it.

Edited to add:

The most obvious effect of having too much copper is gastrointestinal distress. So if you try this and have stomach problems, maybe stop.

Comment by benwr on Coronavirus: Justified Practical Advice Thread · 2020-03-01T08:00:29.979Z · LW · GW

My plan is to use a reusable p100 mask, with copper tape on the structural components, hopefully forming a barrier between the external filters/valves and the internal surfaces, and then to wash my hands before / after wearing it. No idea whether this will be any good, but it seemed more workable / less potentially blameworthy than acquiring disposable masks in a shortage.

Comment by benwr on Coronavirus: Justified Practical Advice Thread · 2020-03-01T01:32:51.583Z · LW · GW

Recommendation: cover the back of your smartphone in copper tape.

Reasoning: in addition to the reasoning for putting copper on all commonly touched surfaces, your phone is an especially good choice because of the "copper halo effect", in which copper ions move from copper surfaces to nearby surfaces (like, say, your hands), leaving them much less hospitable to microbes. [Edit: As MalcolmOcean points out, this mechanism for the halo effect isn't supported by the Wikipedia page below. I made a leap to this explanation without realizing it. That said, I do think that copper (and copper oxides) will get on your hands as a result of this tape, partly because I've seen my hands turning a bit blue.]

A downside is that your hands may turn slightly blue. [Also: See the comment below about uncertainty about how much copper you'll eat as a result of this; tldr: I don't know but I think it's probably fine]

[This is a signal boost for Lady Jade Beacham's response to Connor_Flexman, but the idea was originally introduced to me by James Payor]

Comment by benwr on [deleted post] 2020-01-27T07:27:10.330Z

I think the main difference between the marginal reversal test and how I read your post is just the magnitude of the change. For the marginal reversal test to make sense, I think the change needs to be small relative to "typical" values of the parameter. So, changing life spans by months rather than years, or changing body temperature by single degrees.

And yeah, I think that the counterfactual reversal test is much more of a heuristic than a careful argument, but it does seem useful as a way of disentangling disagreements, especially with sufficiently thoughtful interlocutors.

Comment by benwr on [deleted post] 2020-01-27T07:22:52.133Z

I meant the first thing, sorry for lack of clarity

Comment by benwr on [deleted post] 2020-01-25T19:41:59.749Z

I think that neither of your examples is correctly using the reversal test. IMO, two different versions of the reversal test are useful: the marginal reversal test, and the counterfactual reversal test.

Marginal version: "So you don't think that increasing your body temperature a small amount is good - do you think decreasing it a small amount would be good? If not, can you explain why your current state is optimal?"

Counterfactual version: "So you don't think that increasing your body temperature by 50 degrees would be good? Would you still think that if your body temperature had always been 50 degrees higher, and we were talking about decreasing it 50 degrees?"

I think in both cases the reversal tester correctly loses the argument. I also think that they both do a good job of helping to decide where to find the interesting bits of argument.

Comment by benwr on benwr's unpolished thoughts · 2019-12-04T09:23:30.802Z · LW · GW

I made a blog because I didn’t know where else to write gushing reviews of things. I haven’t written anything there yet, but soon I hope to have written up some of the following:

  • An account of what I’ve learned since getting mildly fixated on pumping CO2 out of my bedroom
  • A gushing review of my iPad Pro 11” with Apple Pencil
  • A mostly-gushing review of my Subaru Crosstrek
  • A gushing review of my bed and mattress
  • A gushing review of the-general-excellence-of-fast-food-franchises
  • A post about how I feel a lot of internal tension about consumerism
Comment by benwr on benwr's unpolished thoughts · 2019-12-04T09:17:25.016Z · LW · GW

I’ve recently been thinking a lot about pumping CO2 out of my bedroom. By coincidence, so has Diffractor (in a slightly different context / with different goals). His post on the CO2 scrubber he built is a pretty good read, although I think he might be making a mistake about the plausibility of vacuum swing adsorption using zeolites. I wrote a comment outlining what I think is the mistake, and I guess I wanted to highlight it here in case I later want to come back and find it, and because I want more people to see it and potentially write dissenting opinions.

Comment by benwr on CO2 Stripper Postmortem Thoughts · 2019-12-04T08:44:35.305Z · LW · GW

Because the air you breathe out has much more CO2 by volume than the air you're processing (it diffuses quickly). See my comment elsewhere for a potential way around this, though.

Comment by benwr on CO2 Stripper Postmortem Thoughts · 2019-12-04T00:59:43.428Z · LW · GW

Adsorption onto zeolites is also plausible, but the issue is that it requires alternately exposing the zeolites to high air pressure and low air pressure, and high airflow is required. We can idealize a perfect CO2 stripper as a magic box that inhales air and spits it out at 0 ppm. If you want a steady-state concentration of 500 ppm for 2 people, then we can see how much air-flow is required to lock up 2 kg of CO2 in 24 hours. This comes out to about 100 cubic feet per minute. This is the bare minimum air flow for any CO2 stripper, but in this particular case, it corresponds to a 25 horsepower air compressor, which is 18 kilowatts.

I'm not sure this is as much of a barrier as it sounds, at least if you have access to a window you can vent to. Imagine if I had a membrane that let only CO2 through, and a 50-micron vacuum pump (typically 1/2 HP). I could pump CO2 out of the room about as fast as the CO2 would diffuse across the membrane. In this setup, the amount of gas pulled out would be pretty tiny, because you'd be pumping only the CO2, but the effective amount of air being processed would be quite large.

So if you have a vacuum swing adsorption machine, with some zeolites in an array of mixing chambers, that can be alternately mixed with the air in the room and then evacuated outside, what matters is the equilibrium adsorption of each component of air, and the speed of adsorption / desorption. The air mixing can be done with a simple fan (providing the effective high volumes of air processing).

That said, there are other major problems with zeolites. The main one I've been struggling with is that zeolites really love to adsorb water vapor, nearly as much as CO2, and there’s a lot more water vapor than CO2 in the air. Competition for adsorption sites isn’t well-understood, but one study shows that water vapor in the air seriously decreases the amount of CO2 adsorbed. This also means that the system described above would function as a dehumidifier as much as a CO2 pump.

[epistemic status: very uncertain; writing as though I were more certain because I think it's more fun / engaging]

Comment by benwr on The best of the www, in my opinion · 2019-10-18T21:49:27.020Z · LW · GW

My list is similar but also includes

Comment by benwr on benwr's unpolished thoughts · 2019-09-08T18:06:46.040Z · LW · GW

Doom circles seem hard to do outside of CFAR workshops: If I just pick the ~7 people who I most want to be in my doom circle, this might be the best doom circle for me, but it won't be the best doom circle for them, since they will mostly not know each other very well.

So you might think that doing doom "circles" one-on-one would be best. But doom circles also have a sort of ceremony / spacing / high-cost-ness to them that cuts the other way: More people means more "weight" or something. And there are probably other considerations determining the optimal size.

So if you wanted to have a not-at-the-end-of-a-workshop doom circle, should you find the largest clique with some minimum relationship strength in your social graph?

Comment by benwr on benwr's unpolished thoughts · 2019-09-07T00:56:55.934Z · LW · GW

Yet another Shortform-as-feature-request:

Notifications and/or RSS feeds from particular posts' comments / answers.

This would be especially useful for Questions and Shortform posts (sometimes tellingly mis-labeled "shortform feeds"), both of which are things where one particular post has a collection of related comments, and which gather content over time.

I currently subscribe to the front page in Feedly, and whenever someone asks a question that I find interesting I mentally cringe because I know that I'll have to remind myself to check back (and I probably will never actually check back).

I guess I could come up with some custom Zapier / IFTTT system for this if I spent a few hours on it, but I suspect this would be generally useful functionality.

Comment by benwr on September Bragging Thread · 2019-09-03T23:38:43.326Z · LW · GW

Update: our application was approved! The Lodge is almost certainly moving!

Comment by benwr on benwr's unpolished thoughts · 2019-08-31T20:07:16.227Z · LW · GW

One friend pointed out that you might be able to avoid some of the pitfalls by releasing something like an open source desktop application that requires you to feed it a database of information. Then you could build databases like this in lots of different ways, including anonymous ones or crowdsourced ones. And in this case it might become a lot harder to claim that the creator of the application is liable for anything. I might actually want to talk to a lawyer about this kind of thing, if the lawyer was willing to put on a sort of "engineering" mindset to help me figure out how you might make this happen without getting sued. So if you know anyone like that, I'd be pretty interested

Comment by benwr on benwr's unpolished thoughts · 2019-08-31T00:20:40.511Z · LW · GW

There should really be a system that does what WebMD tried to do, but actually does it well.

You'd put in your symptoms and background info (e.g. what country you live in, your age), it would ask you clarifying questions ("how bad is the pain from 1 to 10?" "which of these patterns is most like the rash?" "Do you have a family history of heart disease?") and then it would give you a posterior distribution over possible conditions, and a guess about whether you should go to the emergency room or whatever.

Is this just much harder than I'm imagining it would be? It seems like the kind of thing where you could harvest likelihood ratios and put them all into a big database. Is there some regulatory thing where you can't practically offer this service because it's illegal to give medical advice or something?

Comment by benwr on September Bragging Thread · 2019-08-30T23:56:55.953Z · LW · GW

I got 4 out of six people in my group house, plus two others, to apply for a lease in a new and probably-better house! In order to do this I spent probably forty hours in the last month, searching rental listings, reaching out to people, and trying to accommodate people's often-contradictory preferences.

At one point I wrote a program to determine, given a collection of people larger than the number of rooms, and their maximum prices for each room in a house (which are allowed to be conditional on who else is in the house), whether any subset of that group would be able to afford the house. This script was about a hundred lines of python, and after fixing one typo it ran perfectly on the first try. This happens to me pretty often these days when I write python code, and I'm fucking pleased about it. Like, maybe ten years ago I set out to be a master wizard in one particular specialization ("writing programs to solve practical problems") and yeah, I can just really do that kind of magic now. It just feels like "part of me" in a really exhilarating way.

Comment by benwr on jp's Shortform · 2019-08-30T22:32:45.792Z · LW · GW

This distinction is still relevant and useful in some contexts.

Comment by benwr on benwr's unpolished thoughts · 2019-08-05T22:59:51.770Z · LW · GW

As long as I'm using shortform posts to make feature requests, it would be really useful to me to have access to a feed (of shortform posts, normal posts, or both) where I could select which users I see. Right now I come to LessWrong and have a hard time deciding which posts I care about - lots of people here have lots of interests and lots of different standards for content quality, some of which I find actively annoying. Allowing me to build feeds from custom lists of selected users would let me filter by both shared interests and how valuable I typically find those users' posts. I don't think "who is in my custom feed" should be public like it is on Facebook, but Facebook circa 2012 gave me a lot of control over this via friend lists, and I miss the days when that feature was prioritized.

I also like the way solves this problem for some users, although I think the solution above would be better for me personally: every post comes with a selection from a group of site-wide tags, and users can filter their home page based on which tags they want to see.

Comment by benwr on benwr's unpolished thoughts · 2019-07-31T18:55:50.298Z · LW · GW

Interesting, good to know. I'm curious if you considered doing something like, where the avatar is next to the username and the same height as the text.

Comment by benwr on benwr's unpolished thoughts · 2019-07-31T18:05:58.226Z · LW · GW

I should have posted this comment here and rephrased it, sorry:

I also note that avatars could use tricks to solve various constraints I'm imagining the LessWrong team might want to impose.

For example, if you think avatars might make the comments section too visually interesting you could render them in greyscale, or with muted colors. And if you think they might lead to people playing weird games with their avatars (I don't think this is likely, but I can imagine someone worrying about it), you could let users choose from a small collection of acceptable-to-you, auto-generated images based on a hash of their username.

Comment by benwr on benwr's unpolished thoughts · 2019-07-31T18:02:23.191Z · LW · GW

Edit: I don't retract this comment but I should have rephrased it and posted it as a reply to this comment

I also note that avatars could use tricks to solve various constraints I'm imagining the LessWrong team might want to impose.

For example, if you think avatars might make the comments section too visually interesting you could render them in greyscale, or with muted colors. And if you think they might lead to people playing weird games with their avatars (I don't think this is likely, but I can imagine someone worrying about it), you could let users choose from a small collection of acceptable-to-you, auto-generated images based on a hash of their username.

Comment by benwr on benwr's unpolished thoughts · 2019-07-29T02:20:51.283Z · LW · GW

I really think LessWrong would benefit by giving users avatars. I think this would make the site much more visually appealing, but I also think it would vastly decrease the cognitive load required to read threaded conversations.

Comment by benwr on benwr's unpolished thoughts · 2019-07-29T02:18:14.631Z · LW · GW

A while ago, Duncan Sabien wrote a Facebook post about a thing he called "aliveness", and presented it on a single spectrum with something called "chillness". At the time I felt that aliveness seemed sort of like obviously-the-good-one, and like I was obviously-bad-for-being-more-chill, and I felt sad because I think there were a lot of pressures when I was younger to optimize for chillness.

But recently I've been in a couple of scenarios that have changed my views on this. I now think that aliveness and chillness aren't quite opposite ends of the same axis in person space. It seems instead like they're anticorrelated features of a given person in a given situation, and many people live their lives with a nearly-fixed level of each. But there are also people who can control their levels of aliveness or chillness, as the situation demands.

And it isn't the case that chillness is worse. I think it is much, much easier to coordinate large groups of chill people than not-chill people. I think that these people can also definitely be "alive" in the relevant way.

My intuitive feeling is that this ability to control your chillness and aliveness is strongly related to "leadership qualities". And, at least for me, noticing that these might not be opposite ends of a fixed spectrum makes me feel a lot more hopeful about the possibility to grow in aliveness-capability.

Comment by benwr on How/would you want to consume shortform posts? · 2019-07-26T01:19:30.046Z · LW · GW

I'd like to be able to subscribe to individual Shortform feeds via RSS.

Comment by benwr on Why I've started using NoScript · 2019-05-15T23:56:06.089Z · LW · GW

You're right; I'm sorry that I didn't read your comment sufficiently carefully.

The reasoning there is purely my expectation and isn't based on data or particular experience.

Comment by benwr on Why I've started using NoScript · 2019-05-15T23:38:12.990Z · LW · GW

This is a great response and I'm glad to have read it. However I think you miss one important disadvantage of your approach: These alternatives are mostly blacklists, and so they become less useful as you get further into the less-trafficked corners of the web, which is also where you're most likely to hit, e.g., invisible compromised resources.

I've also been surprised at how little "whitelist fatigue" I've gotten. I would have naively expected to get tired of whitelisting domains, but in practice it's continued to feel freeing rather than obnoxiously attention consuming, and site functionality is almost always easy / obvious to enable properly. It's possible that sometimes I miss intended functionality, but I doubt that this comes close to outweighing the benefits.

Edit: the following paragraph misunderstands Said's comment and doesn't address the point that it was meant to; apologies.

Finally, I don't buy the argument about incentivizing web authors. If trackers work less well, there is obviously less incentive to use them. If the only thing holding back authors from adding trackers willy-nilly is user annoyance at page bloat, then it's clearly not enough, and so telling people to just go on shouldering that annoyance to ensure that the annoyance is minimized seems like privileging second-order effects that I would expect to be small.

Comment by benwr on Overconfident talking down, humble or hostile talking up · 2018-11-30T17:19:55.308Z · LW · GW

This nicely explains why I feel so embarrassed when I learn that someone I'm talking with is more knowledgeable than I thought. I wonder how to avoid subconscious overconfidence- / humility-projecting.

It might work to add a TAP for thinking "if this person were much more/less knowledgeable than me, would I have the same presentation in this conversation?"

Comment by benwr on RSS Feeds are fixed and should be properly functional this time · 2018-03-02T10:00:02.398Z · LW · GW

Ah, excellent. I didn't know there was a github.

Comment by benwr on RSS Feeds are fixed and should be properly functional this time · 2018-02-28T21:33:03.694Z · LW · GW

As I mentioned in a post that has now fallen off the end of meta: while RSS feeds work much better now, usernames still don't show up as the author in the feeds.

Comment by benwr on Lesswrong 2016 Survey · 2016-03-26T08:33:27.336Z · LW · GW

Great survey!

However, when you save your progress and are asked to save a password, there's no indication that it will be sent to you in an email or saved at all in recoverable form. I used my least-secure password generation algorithm anyway, but: Do you think you could add a note to the effect that users should not use passwords that they use elsewhere?

Comment by benwr on 'Dumb' AI observes and manipulates controllers · 2015-01-14T02:34:54.189Z · LW · GW

The problem with these particular extensions is that they don't sound plausible for this type of AI. In my opinion it would be easier when talking with designers to switch from this example to a slightly more sci-fi example.

The leap is between the obvious "it's 'manipulating' its editors by recognizing simple patterns in their behavior" to "it's manipulating its editors by correctly interpreting the causes underlying their behavior."

Much easier to extend in the other direction first: "Now imagine that it's not an article-writer, but a science officer aboard the commercial spacecraft Nostromo..."

Comment by benwr on The guardian article on longevity research [link] · 2015-01-12T10:34:38.264Z · LW · GW

On the other hand, the number of people working on a problem, and the speed with which they are individually able to work, can't be ignored. "Given enough eyes, all bugs are shallow" - Linus Torvalds, talking about something pretty similar (if much, much simpler).