All of benkuhn's Comments + Replies

Oops—I forgot that this was going to be auto-crossposted to LW and probably would have prevented that if I remembered, since it's weirdly meta (it was intended mostly as a placeholder for people visiting my website and wondering why there were no recent posts). I guess I'll leave it here now since it got a lot more upvotes than I expected though :)

I think it's still important to note that "not giving up" can lead not just to lack of success, but also to value destruction (Pets.com; Theranos; WeWork). 

If you're going to interpret the original "don't give up" advice so literally and blindly that "no matter what the challenges are I'm going to figure them out" includes committing massive fraud, then yes, it will be bad advice for you. That's a really remarkably uncharitable interpretation.

Not sure if this is your typo or a LW bug, but "essay" appears not to actually be hyperlinked?

I don't think founder/investor class conflict makes that much sense as an explanation for that. It's easy to imagine a world in which investors wanted their money returned when the team updates downwards on their likelihood of success. (In fact, that sometimes happens! I don't know whether Sam would do that but my guess is only if the founders want to give up.)

I also don't think at least Sam glorifies pivots or ignores opportunity cost. For instance the first lecture from his startup course:

And pivots are supposed to be great, the more pivots the better. S

Oops, thanks! Added a link to the signup form on my site. (And fixed my RSS rendering to not do forms like that in the future.)

Yes, and I think the different words were useful!

You're repeating / elaborating on things that are in the post, but were not particularly emphasized. I didn't emphasize them because I've personally had the "deeply internalized felt sense of how easy it is for humans to misunderstand each-other" that you describe for a long time, and only more recently got the "be curious" part, and so I emphasized that because it was the missing piece for me (and didn't totally realize the degree to which the other part was load-bearing / could be the missing piece for others).

Yeah, if you don't want to DIY it, you can apparently also use a teleprompter for a similar effect. I haven't tried one, but am curious to! The only trade-off I can think of (other than cost and being cumbersome) is that you probably sacrifice some image quality from all the reflection shenanigans, but not sure how big of a deal that would be.

1. Even if decent, I'd be surprised if the microphone compares well to the BoomPro (it's farther from your mouth so will pick up more noise, and most non-standalone mics are optimized for low cost not quality)
2. I think most earbuds are a lot worse than Pixel USB-C buds for calls—it looks like these have a relatively nice microphone, and relatively poor noise isolation (= allow you to hear your own voice much better)

Oh OK, then the audio improvements the post describes will work fine as-is (except for the non-headset mic) and the video ones are probably not worth it. And I guess the networking advice becomes "invest in properly debugging your wifi" instead of running a cable.

I assume you don't care about video if you'll be walking around? (I think you lose a lot by giving up video, but de gustibus non disputandum.)

For audio, you could possibly wireless-ify the audio setup in the article with something like an AptX Low Latency to 3.5mm adapter and switching from the BoomPro to the Antlion ModMic Wireless (since I don't think those adapters support mics).

Note that Bluetooth sucks and I haven't tested this so it may not work as well as it sounds like it should. But if you want to keep open-back headphones and a good mic, that's probably your best bet.

Yay, happy to hear it was helpful!

FYI, you can also disable the red circle from within the Slack preferences (maybe you already knew this, but if not, sorry the post wasn't more explicit!)

Thanks, fixed!

I'm confused about the "because I could not stop for death" example. You cite it as an example of GPT-3 developing "the sense of going somewhere, at least on the topic level," but it seems to have just memorized the Dickinson poem word for word; the completion looks identical to the original poem except for some punctuation.

(To be fair to GPT-3, I also never remember where Dickinson puts her em dashes.)

Grubhub used to be over 50% but is now behind Doordash, so maybe doesn't qualify.

Apple has a monopoly on iOS app distribution (aside from rooted phones) and is using it to extract rents, which is what the link is about.

Firefox has 4% market share compared to Chrome's 65%.

Amazon has 40-50% of the ecommerce market depending on which stats you trust.

Google Search has 85%+ market share.

https://www.ftc.gov/tips-advice/competition-guidance/guide-antitrust-laws/single-firm-conduct/monopolization-defined

Courts do not require a literal monopoly before applying rules for single firm conduct; that term is used as shorthand for a firm with significant and durable market power — that is, the long term ability to raise price or exclude competitors. That is how that term is used here: a "monopolist" is a firm with significant and durable market power. Courts look at the firm's market share, but typically do not find

Oops this was super unclear, sorry—the thing that ties together all of these crappy websites isn't money issues, just that they're the winner in a network-effect-based business, thus have no plausible competitors and no incentive to become more useful / less crappy.

What is a "real average"?

I have almost no discipline, I've just spent a lot of time making my habits take so little effort that that doesn't matter :) Figuring out how to make it easy for myself to prioritize, and stick to those priorities, every day is actually a common recurring weekly review topic!

(I considered laying out my particular set of todo-related habits, but I don't think they'd be very helpful to anyone else because of how personal it is—the important part is thinking about it a lot from the perspective of "how can I turn this into a habit that doesn't require discipline for me," not whatever idiosyncratic system you end up with.)

Thanks! Glad you enjoyed them!

Thanks, this comment is really useful!

It is generally accepted that you do not need to go to direct sunlight type lux levels indoors to get most of the benefits.... I am not an expert on the latest studies, but if you want to build an indoor experimental setup to get to the bottom of what you really like, my feeling is that installing more than 4000 lux, as a peak capacity in selected areas, would definitely be a waste of money and resources.

Do you have any pointers to where I might go to read the latest studies?

A typical 60W equivalent LED bulb draws 7.5W and is 90% efficient

Where are you getting this number? As far as I know, the most efficient LEDs today are around 50% efficient.

I've done a bit of research on this. I think something along these lines is practical. My biggest uncertainty is what a "usable form factor" is (in particular I don't know how much diffusion you'd need, or in what shape, with a very small emitter like this).

FWIW, the Yuji chips are insanely expensive per lumen and seem to be on the low end of efficiency (actually they seem like such a bad deal that I'm worried I'm missing something). The chip that came out on top in my spreadsheet was this Bridgelux chip which is about 1/... (read more)

I'm one of the friends mentioned. Here's some more anecdata, most importantly including what I think is the current easiest way to try a lumenator (requiring only one fixture instead of huge numbers of bulbs):

I don't have seasonal depression, but after spending a winter in a tropical country, it was extremely noticeable that it's harder for me to focus and I have less willpower when it's dark out (which now starts at 4:15). I bought an extremely bright light and put it right next to my desk, in my peripheral vision while I work. It... (read more)

Every time I pay for electricity for my computer rather than sending the money to a third world peasant is, according to EA, a failure to maximize utility.

I'm sad that people still think EAers endorse such a naive and short-time-horizon type of optimizing utility. It would obviously not optimize any reasonable utility function over a reasonable timeframe for you to stop paying for electricity for your computer.

More generally, I think most EAers have a much more sophisticated understanding of their values, and the psychology of optimizing them, than you ... (read more)

Yes, I glossed over the possibility of prisons bribing judges to screw up the data set. That's because the extremely small influence of marginal data points and the cost of bribing judges would make such a strategy incredibly expensive.

Yep. Concretely, if you take one year to decide that each negative reform has been negative, the 20-80 trade that the OP posts is a net positive to society if you expect the improvement to stay around for 4 years.

To increase p'-p, prisons need to incarcerate prisoners which are less prone to recidivism than predicted. Given that past criminality is an excellent predictor of future criminality, this leads to a perverse incentive towards incarcerating those who were unfairly convicted (wrongly convicted innocents or over-convinced lesser offenders).

If past criminality is a predictor of future criminality, then it should be included in the state's predictive model of recidivism, which would fix the predictions. The actual perverse incentive here is for the prisons ... (read more)

Gwern has a point that it's pretty trivial to run this robustness check yourself if you're worried. I ran it. Changing the $1 to $100 reduces the coefficient of EA from about 1.8 to 1.0 (1.3 sigma), and moving to $1000 reduces it from 1.0 to 0.5 (about two sigma). The coefficient remains highly significant in all cases, and in fact becomes more significant with the higher constant in the log.

What do you mean by "dollar amounts become linear"? I haven't seen a random variable referred to as "linear" before (on its own, without reference to another variable as in "y is linear in x").

For people who would otherwise not have multiple credit cards, the increase in credit score can be fairly substantial.

In addition to Dorikka's comment, you are not liable for fraudulent charges; usually the intermediating bank is.

If you don't want to bother signing up for a bunch of cards, the US Bank Cash+ card gives 5% cash back for charitable donations, up to I think $2000 per quarter. This is a worse percentage but lower-effort and does not ding your credit (as long as you don't miss payments, obvs).

Also, as I understand, it's actually better not to cancel the cards you sign up for (unless they have an annual fee), because "average age of credit line" is a factor in the FICO score. Snip them up, set up auto-pay and fraud alerts and forget about them, but don't cancel them.

Of course, the case of Beanie Babies is more comparable to Dogecoin than Bitcoin, and the Dutch tulip story has in reality been quite significantly overblown (see http://en.wikipedia.org/wiki/Tulip_mania#Modern_views , scrolling down to "Legal Changes"). But then I suppose the reference class of "highly unique things" will necessarily include things each of which has unique properties... :)

I think the way to go here is to assemble a larger set of potentially comparable cases. If you keep finding yourself citing different idiosyncrati... (read more)

The difference is that it's easy to make more tulips or Beanie Babies, but the maximum number of Bitcoins is fixed.

Yes, this is what I mean by reference class tennis :)

Actually, according to Wikipedia, it's hypothesized that part of the reason that tulip prices rose as quickly as they did was that it took 7-12 years to grow new tulip bulbs (and many new bulb varieties had only a few bulbs in existence). And the Beanie Baby supply was controlled by a single company. So the lines are not that sharp here, though I agree they exist.

Is my general line of reasoning correct here, and is the style of reasoning a good style in the general case? I am aware that Eliezer raises points against "small probability multiplied by high impact" reasoning, but the fact is that a rational agent has to have a belief about the probability of any event, and inaction is itself a form of action that could be costly due to missing out on everything; privileging inaction is a good heuristic but only a moderately strong one.

Sometimes, especially in markets and other adversarial situations, inac... (read more)

I was told that you only run into severe problems with model accuracy if the base rates are far from 50%. Accuracy feels pretty interpretable and meaningful here as the base rates are 30%-50%.

It depends on how much signal there is in your data. If the base rate is 60%, but there's so little signal in the data that the Bayes-optimal predictions only vary between 55% and 65%, then even a perfect model isn't going to do any better than chance on accuracy. Meanwhile the perfect model will have a poor AUC but at least one that is significantly different from... (read more)

I would beware the opinions of individual people on this, as I don't believe it's a very settled question. For instance, my favorite textbook author, Prof. Frank Harrell, thinks 22k is "just barely large enough to do split-sample validation." The adequacy of leave-one-out versus 10-fold depends on your available computational power as well as your sample size. 200 seems certainly not enough to hold out 30% as a test set; there's way too much variance.

Is this ambiguous?

It wasn't clear that this applied to the statement "we couldn't improve on using these" (mainly because I forgot you weren't considering interactions).

I excluded the rater and ratee from the averages.

Okay, that gets rid of most of my worries. I'm not sure it account for covariance between correlation estimates of different averages, so I'd be interested in seeing some bootstrapped confidence intervals). But perhaps I'm preempting future posts.

Also, thinking about it more, you point out a number of differences between corr... (read more)

Nice writeup! A couple comments:

If the dataset contained information on a sufficiently large number of dates for each participant, we could not improve on using [frequency with which members of the opposite sex expressed to see them again, and the frequency with which the participant expressed interest in seeing members of the opposite sex again].

I don't think this is true. Consider the following model:

• There is only one feature, eye color. The population is split 50-50 between brown and blue eyes. People want to date other people iff they are of the

I can't speak to "best," but I suggest reading Style: Lessons in Clarity and Grace by Joseph M. Williams, which crystallizes lots of non-trivial components of "good writing." (The link is to an older, less expensive edition which I used.)

I'll also second "write a lot" and "read a lot." Reading closely and with purpose in mind will speed up the latter (as opposed to the default of throwing books at your brain and hoping to pick up good writing by osmosis). Also, read good writers.

In your "critiquing bias" section you allege that 3/43 studies supporting a link is "still surprisingly low". This is wrong; it is actually surprisingly high. If B ~ Binom(43, 0.05), then P(B > 2) ~= 0.36.*

*As calculated by the following Python code:

from scipy.stats import binom
b = binom(43, 0.05)
p_less_than_3 = sum(b.pmf(i) for i in [0,1,2])
print 1 - p_less_than_3

I think you're being a little uncharitable to people who promote interventions that seem positional (e.g. greater educational attainment). It may be true that college degrees are purely for signalling and hence positional goods, but:

(a) it improves aggregate welfare for people to be able to send costly signals, so we shouldn't just get rid of college degrees;

(b) if an intervention improves college graduation rate, it (hopefully) is not doing this by handing out free diplomas, but rather by effecting some change in the subjects that makes them more capable ... (read more)

Fun question.

The takeover vector that leaps to mind is remote code execution vulnerabilities on websites connected to important/sensitive systems. This lets you bootstrap from ability to make HTTP GET requests, to (partial) control over any number of fun targets, like banks or Amazon's shipping.

The things that are one degree away from those (via e.g. an infected thumb drive) are even more exciting:

• Iranian nuclear centrifuges
• US nuclear centrifuges
• the electrical grid
• hopefully not actual US nuclear weapons, but this should be investigated...

Plausible f... (read more)

Yes, definitely agree that politicians can dupe people into hiring them. Just wanted to raise the point that it's very workplace-dependent. The takeaway is probably "investigate your own corporate environment and figure out whether doing your job well is actually rewarded, because it may not be".

I'd beware conflating "interpersonal skills" with "playing politics." For CEO at least (and probably CTO as well), there are other important factors in job performance than raw engineering talent. The subtext of your comment is that the companies you mention were somehow duped into promoting these bad engineers to executive roles, but they might have just decided that their CEO/CTO needed to be good at managing or recruiting or negotiating, and the star engineer team lead didn't have those skills.

Second, I think that the "playing p... (read more)

I would expect the relevant factor to be mental, not physical, exertion. Unfortunately that's a lot harder to measure.

Do you have actual data on this? Otherwise I'm very tempted to call typical mind.

One story for exponential growth that I don't see you address (though I didn't read the whole post, so forgive me if I'm wrong) is the possibility of multiplicative costs. For example, perhaps genetic sequencing would be a good case study? There seem to be a lot of multiplicative factors there: amount of coverage, time to get one round of coverage, amount of DNA you need to get one round of coverage, ease of extracting/preparing DNA, error probability... With enough such multiplicative factors, you'll get exponential growth in megabases per dollar by applying the same amount of improvement to each factor sequentially (whereas if the factors were additive you'd get linear improvement).

If anyone's admitted/visiting Harvard, let me know! I go there and would be happy to meet up and/or answer your questions. There are some other students on here as well.

"outstanding" still has some of the same connotations to me, although less so. But I may be in the minority here.