Dan Luu

Public Key

npub1pm93exm0eyfep2635k3epqh5f39tuzse44nuvwmsn5t4a0erureqnl88nl

NIP-05 Address

Profile Code

nprofile1qqsqajcundhujyus4dg6tgusst6ycj47pgv66e7x8dcf6967hu37puspz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dstr3p4q

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1pm93exm0eyfep2635k3epqh5f39tuzse44nuvwmsn5t4a0erureqnl88nl

Show more details

Last Notes

2026-05-22 08:06:16 GMT

Why are so many people so sure that the big AI providers are losing money on inference? It reminds me of the comments about how Uber can never make money. Their unit economics were fine and they were only losing money because they chose to do so on customer acquisition.
In this case, I don't think it's even obvious that inference should be money losing today. API cost for frontier models has gone way up and the people claiming some cost for inference are relying on made up assumptions.
https://files.mastodon.social/media_attachments/files/116/617/165/666/552/364/original/607d038ab7db9c7a.png

2026-03-30 08:00:29 GMT

Interesting to see Copilot injecting ads into PR descriptions. Although there are a handful of older instances of this, if GitHub search is working properly, it looks like this started happening at scale around 10 days ago with more than 1k injections of this particular ad per day since then
https://github.com/search?q=%22%E2%9A%A1+Quickly+spin+up+copilot+coding+tasks+from+anywhere+on+your+macOS+or+Windows+machine+with+Raycast%22&type=pullrequests&s=created&o=asc&p=1
What will they think of next?
https://files.mastodon.social/media_attachments/files/116/316/995/876/013/772/original/0e1d6fe804a155f7.png
https://files.mastodon.social/media_attachments/files/116/316/996/264/770/284/original/b4a07509e1cdc8b7.png

2025-05-26 05:46:39 GMT

Interesting story about Google publishing someone's phone number when they gave it to Google for account verification/security:
https://danq.me/2025/05/21/google-shared-my-phone-number/
Reminds me of the time a company I worked for (AFAIK) accidentally used phone numbers obtained the same way for ad targeting and got fined $150M. Scaled by company size, the equivalent find would be ~$10B.
https://files.mastodon.social/media_attachments/files/114/572/548/187/932/591/original/76550031a16dde86.png

2025-04-02 22:23:53 GMT

Is one of the big drivers of consumer demand for compute going to be solving expensive proof of work proxies?
I'm still semi-banned from reddit (https://mastodon.social/@danluu/113426851280402188) and, ironically, browse it with a scraper because the scraper has good anti-blocking techniques and I don't, much like how way SEO spammers are better at ranking than people who make legit content.
With the resources AI companies have, proof of work can't "win", but it seems like the least ineffective technique?
https://files.mastodon.social/media_attachments/files/114/270/687/541/712/235/original/8c116e0673be2c0f.png

2024-11-22 22:41:22 GMT

Commodore 64 version of Missile Command where the bottom of your screen is the game state in memory and missiles cause memory corruption, which eventually causes you to lose: https://csdb.dk/release/?id=135463.
In the video below, a missile broke my controls and caused my cursor to move down and to the left so I couldn't stop other missiles.
https://files.mastodon.social/media_attachments/files/113/528/994/760/698/312/original/89af61339f14e57b.mp4

2024-10-12 06:10:34 GMT

I find it very funny that "poaching" is considered inherently unethical or immoral in some circles.
There's a blog post I've thought about writing titled something like "progressives in favor of the 0.1%" that discusses norms and laws that my lefty social circles advocate for that mostly benefit the extremely wealthy.
Who doesn't like poaching? Eric Schmidt and Steve Jobs (https://danluu.com/google-wage-fixing/), business owners and capital, and people who claim to be looking out for the little guy.
https://files.mastodon.social/media_attachments/files/113/292/929/878/736/518/original/e10f47beff85d05b.webp
https://files.mastodon.social/media_attachments/files/113/292/930/224/616/526/original/a79789972ed24394.webp

2024-08-12 04:50:57 GMT

How good can you be at Codenames without knowing any words?
https://danluu.com/codenames/
https://files.mastodon.social/media_attachments/files/112/947/241/589/126/869/original/3d7be4319c6fd255.mp4

2024-07-22 07:04:39 GMT

Why wasn't Steve Yegge's YouTube channel more popular? There was a fairly long time period where he seemed to be widely read programming blogger, which you might think is a good springboard for doing video, but his median video has something like 2k views.
For people who haven't seen it, the first episode is https://www.youtube.com/watch?v=C8332hz8c2s, and he talks a bit about why he ended up winding down the channel and taking a job in episode 49: https://youtu.be/sKE1S7PK1fY?si=2g2lcrR7Ls0IikVm&t=2977.

2024-07-13 06:23:17 GMT

Although this didn't end up working out in practice (at least in the general case), I still think this paper is super cool:
https://dl.acm.org/doi/pdf/10.1145/237090.237173
https://files.mastodon.social/media_attachments/files/112/777/739/868/980/252/original/46f5596f0edcda41.png

2024-07-07 20:28:20 GMT

What's wrong with Crockford's DEC64 type, which he says is "intended to be the only number type in the next generation of application programming languages"
https://www.patreon.com/posts/107699971
https://files.mastodon.social/media_attachments/files/112/747/090/305/199/117/original/3b33dbae5ce5517f.png

2024-06-26 22:40:55 GMT

On testing this, I was surprised to find that there's a real, measurable, difference between a "good" ($15, Mogami or Canare) XLR cable for my microphone and cheap cables ($8, Amazon Basics or Monoprice) or fancy but not good cables (Rode, $40).
I always thought of "good cables" as audiophile nonsense, but I guess microphones are different in that you're taking the any noise and then giving it 30dB of gain or w/e, plus my mic cable runs past a lot of big noise sources.

2024-06-25 08:00:09 GMT

A lot of things have actually improved quite a bit, or how the monthly "everything sucks" rant is wrong
https://www.patreon.com/posts/106839697
https://files.mastodon.social/media_attachments/files/112/676/200/788/780/267/original/914507cd5c4f1d64.png

2024-06-17 05:55:55 GMT

A discussion of discussions on AI Bias:
https://danluu.com/ai-bias/
https://files.mastodon.social/media_attachments/files/112/630/414/142/178/626/original/ad0d5558c1b439d8.png

2024-06-14 21:16:37 GMT

Do a lot more people use their phones while driving than I think?
Police here announced that they were going to do a distracted driving month where they'd ticket people for using (touching) their phones while driving.
The two times I saw police waving people over for this, they were getting maybe 1/4 to 1/3 of drivers on this straight stretch of road near a school, pulling up to a red light.
Can it really be the case that 1/4-1/3 of drivers are actively on their phones? That seems too high?

2024-06-13 23:38:10 GMT

A brief history of Google Pay:
https://news.ycombinator.com/item?id=39474629
https://files.mastodon.social/media_attachments/files/112/611/940/920/383/429/original/3b42fc47fc02e148.png

2024-06-12 18:44:38 GMT

This is such a classic big company story: someone finds a critical problem (in this case, the team manually labelling training data is mislabelling it), but the team that owns the issue doesn't care about it and no consumer of the team's product has the bandwidth/patience to go through the many months long process of getting the issue fixed, so the issue never gets fixed:
https://www.mihaileric.com/posts/how-alexa-dropped-the-ball-conversational-ai/
https://files.mastodon.social/media_attachments/files/112/605/124/333/883/557/original/492b7db042ffd79d.png

2024-06-08 09:43:24 GMT

- reply

@npub1d36…374v Thanks! That looks like another version of https://mastodon.social/@danluu/112559852135059558.
I think I now know what causes this and have a workaround for it.

2024-06-07 08:53:39 GMT

While I was going through old files, I found this set of forum threads from 2008 on work-life balance at Bioware:
https://danluu.com/bioware/
I originally ran into these after learning about something called "sympathy crunch" from someone at Bioware, who claimed this was common there. A sympathy crunch is where you end up "crunching" even though you don't have any work to do, stuck at the office because other people have work to do.

2024-06-03 02:19:44 GMT

The cost of construction in Vancouver
https://www.patreon.com/posts/105464440
https://files.mastodon.social/media_attachments/files/112/550/287/560/143/296/original/aa7305866cf99413.png

2024-05-30 22:20:48 GMT

- reply

@npub1jr0…zdwr Whoops, I guess I missed one. Maybe I should've asked an AI image editor to do it.

2024-05-30 22:05:05 GMT

The future of support: "I wasn't able to convince the support AI this was urgent enough to grant me access."
https://news.ycombinator.com/item?id=40529033
https://files.mastodon.social/media_attachments/files/112/532/302/457/254/682/original/3ce199af223b349d.png

2024-05-26 21:21:49 GMT

A tech-focused analysis of the 2011-2012 FTC antitrust investigation of Google:
https://danluu.com/ftc-google-antitrust/
https://files.mastodon.social/media_attachments/files/112/509/475/364/759/050/original/fe62a85432a07020.png

2024-05-21 20:00:16 GMT

- reply

BTW, I don't use iTerm2 because, per https://danluu.com/term-latency/, I find the latency to be too high.
But if I were concerned about this and I preferred iTerm2, I would spend 5 minutes writing a diff to remove this feature. It's an open source piece of software and I suspect that you could stub out the feature in a patch in less time than has been spent on discussion about how terrible AI terminal integration is.

2024-05-19 02:28:56 GMT

This study that looks into what errors CS students make (in an optional class mostly taken by 3rd/4th year CS majors, with 10% grad students) is fascinating:
https://cs.brown.edu/~sk/Publications/Papers/Published/lk-smol-tutor/paper.pdf
Among other things, it really shows how reasoning about computer programs is not intuitive to people, even people who've had years of training.
https://files.mastodon.social/media_attachments/files/112/465/385/157/745/666/original/468852d7e0a1724f.png

2024-05-10 17:59:58 GMT

Interesting to see users working around the lack of tactile buttons in some modern cars:
https://files.mastodon.social/media_attachments/files/112/418/086/833/184/364/original/bc2be040bb5a4241.png

2024-04-24 14:27:29 GMT

How long before formal methods are widely use in software engineering?
https://www.patreon.com/posts/102809124
https://files.mastodon.social/media_attachments/files/112/326/657/140/282/793/original/6e0b5275341f9fb2.png

2024-04-17 18:56:26 GMT

I see these AI generated summaries are going great.
BTW, I mean this non-ironically. This is generating a huge amount of engagement and juicing user numbers, which generally companies care more about than accuracy of news.
For people not familiar with basketball, a bad shot that misses is called a "brick" and Klay Thompson put up a lot of bricks against the Sacramento Kings. People talking about this resulted in Twitter creating this AI generated trend.
https://files.mastodon.social/media_attachments/files/112/288/076/890/734/037/original/93fca802a2467251.png

2024-04-15 02:09:24 GMT

- reply

GDP only directly appears in the denominator of Medals/GDP, so the ranking implies that having high GDP is bad.
I don't particularly care about country rankings, but I feel like this same idea applies to benchmarks. The benchmark methodology tells you what the person thinks is important, e.g., when people pass around benchmarks of terminal throughput to prove that some terminal is the best, they're effectively saying that terminal throughput is very important.

2024-04-07 05:05:48 GMT

Interesting comment about SGI leadership knowing about the problems they were facing and still being unable to come up with a way to handle them.
https://files.mastodon.social/media_attachments/files/112/228/190/886/224/119/original/6a11891c2418c879.png

2024-03-24 03:39:17 GMT

These sites that are gaming search ranking by generating nonsense pages that match any query are getting a bit ridiculous (the highlighted result is the 4th highest result for the query on DDG and Bing).
https://files.mastodon.social/media_attachments/files/112/148/579/617/122/801/original/5e14ffa31a0f6421.png

2024-03-16 19:57:17 GMT

How web bloat impacts users with slow devices:
https://danluu.com/slow-device/
https://files.mastodon.social/media_attachments/files/112/107/122/144/005/594/original/63cdb917d9efb4a2.png

2024-03-07 19:53:44 GMT

I'm blown away by how good introductory programming resources are nowadays. I tried out the javascript / regex / sql stuff on executeprogram.com and it's just way better than anything that existed 20 years ago. Likewise for PHP stuff at https://laracasts.com/. I don't love that it's video but, even so, it's much more beginner friendly than a ruby or perl book from 20 years ago.
See also, Julia Evans's work for systems, as well as https://pages.cs.wisc.edu/~remzi/OSTEP/ for a not quite beginner text, etc.

2024-02-25 07:53:36 GMT

For people who do performance measurements on mobile devices via Chrome Dev Tools Android, what do you use to automate this?
https://developer.chrome.com/docs/devtools/remote-debugging is a nice guide on how to manually do this, but what do you do for automation?
I guess I can read the source code for webpagetest, which must have some way to automate this, but is there, documentation, some API name I should look up, or a simple example out there?

2024-02-24 21:09:21 GMT

- reply

On a related note, I found Jonathan Shapiro's retrospective on BitC (which aimed to fill the niche Rust now occupies) interesting:
https://danluu.com/bitc-retrospective/
https://files.mastodon.social/media_attachments/files/111/988/499/244/780/784/original/fa582506a72b4eda.png

2024-02-23 18:43:22 GMT

- reply

@npub1pu3…8r76 It's hard to tell about prevalence from the outside, but I feel like getting a part that hadn't undergone sufficient manufacturing testing used to be extremely rare.
The cases I used to hear about (25 years ago) were things like, you overclock and overvolt your CPU for a year and then it starts being unstable, but those CPUs were never designed to be used that way, whereas now it's expected that many users will get "gaming" (or w/e) boards that do this and the parts won't work.

2024-02-23 17:15:37 GMT

Coming up on a decade since https://danluu.com/cpu-bugs/ and the prediction seems to be holding up: https://www.radgametools.com/oodleintel.htm.
I wonder if we'll ever go back to the level of verification effort Intel used to put in. I suspect not in my lifetime. Historically, AMD was much worse than Intel w.r.t. serious stability bugs and, rather than AMD catching Intel, Intel seems to have caught AMD.
Intel has not caught up to Nvidia in bugginess, but maybe that's something to look forward to in the next decade.
https://files.mastodon.social/media_attachments/files/111/981/866/334/396/606/original/4d13a48b81d77ccb.png

2024-02-19 20:11:04 GMT

- reply

@npub1le5…gagn Whoops, I guess the link got removed! It's https://danluu.com/atom.xml

2024-02-18 20:14:47 GMT

Diseconomies of scale in moderation, support, fraud, and spam:
https://danluu.com/diseconomies-scale/
https://files.mastodon.social/media_attachments/files/111/954/310/422/613/803/original/356f93ae9e76a2ce.png

2024-02-09 21:39:15 GMT

Reports of Microsoft's death are greatly exaggerated
https://www.patreon.com/posts/98126521

2024-02-08 03:21:58 GMT

The impossibility of agreement on what should be allowed:
https://danluu.com/impossible-agree/
https://files.mastodon.social/media_attachments/files/111/893/706/299/612/429/original/9eb149a84eaac707.png

2024-02-02 17:28:03 GMT

In other https://danluu.com/discontinuities/ news, I'm surprised to see that people "round" to nice, even, ages in (some?) age surveys, e.g., see this poll on the age of WordPress users, which has spikes at 25, 30, 40, 45, 50, and 55:
https://wordpress.org/news/files/2023/05/2022-Annual-Survey-Final.pdf
https://files.mastodon.social/media_attachments/files/111/863/037/867/079/831/original/3be771a4c85d655f.png

2024-01-31 23:33:23 GMT

- reply

I know of kafkaesque horror stories of bank and brokerage account loss, so it's not like brokerages are perfect, but it's rare enough that I don't personally know anyone who's had their personal account or funds temporarily lost, let alone permanently, whereas with FB, a decent fraction of my non-tech friends have lost accounts.
In one case, the account over has been running bitcoin scams for maybe a year? And of course FB's anti-fraud systems don't detect this, nor has support dealt with this.

2024-01-31 23:29:18 GMT

- reply

"you can't provide support at scale because it's a social engineering attack vector". This doesn't seem prima facie absurd because telecoms do this and it's supposedly quite easy to just steal someone's phone number and use that to steal people's FB or other accounts.
But at the same time, online brokerages and banks are much higher value targets than FB accounts and yet, people permanently losing their FB account to "hackers" seems to happen at a higher than losing bank/brokerage accounts.

2024-01-31 23:26:18 GMT

Are there fundamental reasons that a company the size of FB can't provide much better support than they do?
The most common explanation I've heard is that support isn't affordable, but I don't find this plausible based the profit FB-sized companies make per user. If you just naively look at how much you could pay support people, it's quite a bit, not including things like diverting money from the ~$50B that's allegedly been spent on the metaverse.
A somewhat more plausible argument I heard is
https://files.mastodon.social/media_attachments/files/111/853/145/649/411/166/original/06ba63e414269277.png

2024-01-31 01:11:53 GMT

Brotli curiosities:
https://www.patreon.com/posts/97523860

2024-01-29 19:51:53 GMT

Notes on Cruise's pedestrian accident:
https://danluu.com/cruise-report/
https://files.mastodon.social/media_attachments/files/111/840/948/550/667/333/original/476eda6d16c934f8.png

2024-01-27 19:27:30 GMT

It's amazing how widely used Teams is considered how widely hated it is: https://news.ycombinator.com/item?id=39157678
It's a true testament to the effectiveness of Microsoft's enterprise sales teams (and strategy).
https://files.mastodon.social/media_attachments/files/111/829/538/365/381/319/original/85b7f8704d24f68a.png
https://files.mastodon.social/media_attachments/files/111/829/538/860/902/570/original/a56d4379a4a8c76c.png
https://files.mastodon.social/media_attachments/files/111/829/539/189/382/376/original/9b169cb910d00d58.png

2024-01-26 22:57:16 GMT

- reply

@npub1wfy…ahvr Interesting. It being an intentional softball question is consistent with the previous "do you plan to keep Facebook free?" as well as with Hatch's not caring about the answer whatsoever and shuffling through his notes for the next question.
And, either way, laughing at a question from a Senator seems like a blunder. It seems a bit odd they didn't practice fielding off-the-wall questions to avoid having the exact reaction they did.

2024-01-26 21:32:26 GMT

It's amazing how little many of the people who set tech policy understand about tech.
E.g., here's Senator Orin Hatch bragging about chairing the Senate High-Tech Task Force and then confusedly asking Zuckerberg how the company can possibly sustain a business model where they don't charge users.
Zuck, caught off guard, takes a long pause and responds, "Senator, we run ads".
Zuck and the two FB staffers behind him (a VP and a lobbyist) hold in their laughter, with varying degrees of success.
https://files.mastodon.social/media_attachments/files/111/824/380/436/952/317/original/5212363fdb142509.mp4

2024-01-25 20:48:00 GMT

Why do people post on [bad platform] instead of [good platform]?
https://danluu.com/why-video/
https://files.mastodon.social/media_attachments/files/111/818/546/899/130/212/original/266141270c55f0c0.png

2024-01-25 01:50:13 GMT

It's pretty incredible to see, in today's fundraising environment, a search engine startup use 1/3 of its fundraising round (and a lot of time and attention) setting up a t-shirt production operation so they can give away free t-shirts *that don't even have the company's name on them*.
It sounds like a story from five years ago.
https://files.mastodon.social/media_attachments/files/111/814/059/562/922/103/original/c21c686997bc43c4.png

2024-01-25 00:05:02 GMT

- reply

@npub1q9s…6tvc Yes, it's possible the author actually means "very fast" and not "exponential".
It's from someone who I wouldn't expect to do that from the other content of theirs I've seen, but I don't know them and don't read all of their stuff, so it's plausible I'm overestimating the precision of their writing here.

2024-01-24 23:51:07 GMT

- reply

that seems extremely unlikely?
2. It seems easy to name an arbitrarily large number of technologies where exponential improvements aren't only or mostly due to semiconductors. To pick an example that most tech-adjacent people wouldn't think of, the (average) YoY improvements in car tires are pretty incredible. While advances in simulation have contributed to advances in tires, advances don't seem to mostly come from simulation?
One question I have is: why would someone think this is true?

2024-01-24 23:46:48 GMT

Is this true? As someone who hasn't seriously studied the history of technology, it doesn't feel true to me.
Two reasons this feels untrue:
1. Before there were semiconductors, exponential improvements in technologies were common. It would be extraordinary if this is now false except when semiconductors were involved. The overwhelmingly likely mechanism for that would be if semiconductors caused such fast improvements in everything that non-semiconductor improvements are a rounding error, but
https://files.mastodon.social/media_attachments/files/111/813/587/319/762/153/original/b5bf68e95ef4105d.png

2024-01-24 22:08:35 GMT

This discussion on the GitHub bug that let you make a "GitHub.com signed" commit as anybody (which is tagged as "Verified") makes me wonder if I'm missing something about parsers vs. regexes.
I'm not an expert (I've written maybe ~10 parsers on my life) but, in my mind, parsers seem easier to test and verify.
Many people seem to think that a parser would've had the exact same bug, but I would expect that a parser with the exact same bug would error out when it encountered an unexpected field.
https://files.mastodon.social/media_attachments/files/111/813/202/490/803/188/original/9ed3723a78799c48.png

2024-01-24 07:52:17 GMT

- reply

At a meta level, something I find funny is that there's a group of people who find the concept of "triggers" and "trigger warnings" to be total bullshit.
That's not an inherently incoherent position, but people in this group seem disproportionately likely to absolutely lose their minds when they see certain things related to tech or engineering, like, furry-style cartoons, the word "decolonization", trigger warnings, etc.

2024-01-24 07:10:48 GMT

- reply

because they saw the word "decolonization". From the comments, people clearly don't know how this word is being used or even what it means in any context, but they know it's definitely bullshit!
If you want to argue that we shouldn't do this, ok, make that argument, but getting so mad when you see a word that you write a rant without finding out what it means seems odd? And likewise for seeing a certain kind of cartoon and then feeling compelled to write a rant about how the site author sucks.
https://files.mastodon.social/media_attachments/files/111/809/672/768/735/067/original/35b61f1d3b61b190.png

2024-01-24 07:04:14 GMT

- reply

The reason the book has been influential is because "decolonization" is this abstract thing a lot of people want to care about, but it's often unclear what it means to do something about it and this book (which is aimed at faculty) provides concrete guidance on how they can change their courses if they want to incorporate decolonization into their classes.
Anyway, 1 out of 16 comments appears to be from someone who opened the book, and the other comments are all from people who lost their minds

2024-01-24 07:01:14 GMT

- reply

Another one is anything related to "decolonization". People just lose their minds when they see the word, e.g., I posted a textbook I think is interesting to HN, Decolonizing the Engineering Curriculum. I know it's been fairly influential at UBC (depending on who you ask, the #2 or #3 ranked university in Canada) and it's changed how at least 15 courses are taught (probably more, that's just what I know of) and it's been used elsewhere as well.

2024-01-24 06:57:08 GMT

I find the set of topics that programmers absolutely lose their minds about absolutely fascinating.
One example is "furry-style" cartoons. As mentioned in https://twitter.com/altluu/status/1479541394884771847, lobste.rs had to ban everyone who loses their mind when they see these cartoons; before the bans every discussion on a post with these kinds of cartoons turned into a flamewar about whether someone can be taken seriously if they have cartoons on their website.
HN also nukes these comments for the same reason.

2024-01-22 23:19:30 GMT

- reply

Long-time Microsoft employees explain changes in Windows:
https://news.ycombinator.com/item?id=30019307
Designers were handed full control over UX. Engineers who fought for usability over a slick-looking interface burned out and left after repeatedly being overruled.
https://files.mastodon.social/media_attachments/files/111/802/153/606/784/287/original/d18eefeb81da16b5.png

2024-01-22 16:05:29 GMT

With a transit strike happening starting today, a bunch of local forums have essentially the same discussion:
A: I'm too poor to Uber and I'm not going to be able to get to work
B: Just walk
C: I'm not going to walk 2/3/4 hours
B: what's the problem? Just walk
C: I'm disabled and/or have a family or other commitments and can't spend 2/3/4 hours commuting each way
D: So? Solidarity
People seem to have a hard time understanding that not everyone is an able-bodied person with infinite free time.
https://files.mastodon.social/media_attachments/files/111/800/442/876/528/204/original/4db8dfb85625245b.png
https://files.mastodon.social/media_attachments/files/111/800/443/136/548/371/original/52d0bd52922d9088.png

2024-01-22 03:03:49 GMT

- reply

I found out that the business had always hemorrhaged money and only "worked" due to trust fund money, which ran out
Based on means, the absolute rate of people doing this kind of thing must be <1% of the U.S. pop, but the relative rate among rich-presenting people seems somewhat high? I must know 10+ who've done this, and that's just the ones who've been discovered
BTW, I'm not counting things like successfully selling scam "get rich" courses, since selling a lot of those is a kind of success

2024-01-22 02:59:06 GMT

In the same vein as https://mastodon.social/@danluu/109537224752779343, something that surprised me is the rate of people LARPing the lifestyle of a successful businessperson, farmer, or whatever.
Now that you can make a ton of money on insta/youtube/etc., there are famous examples of this, but what got me were the mundane examples, e.g., this seemingly successful businessperson would constantly post about her business acumen, and then one day she starts posting about being a realtor, which seemed weird. Years later,
https://files.mastodon.social/media_attachments/files/111/797/355/344/184/159/original/94a50318ea3e63ac.png

2024-01-19 15:38:13 GMT

- reply

@npub1f87…qdlz I don't think this is obvious at all unless we either don't have cities anymore or rich people no longer live in or near cities.
In this world, we have no understand of disease spread or causes of death. We saw this as cities got larger and denser even with some medicine and developing understanding of what was happening and it would be much worse in this counterfactual world.

2024-01-18 04:47:30 GMT

Current and former Google engineers discuss how Google has changed:
https://files.mastodon.social/media_attachments/files/111/775/134/290/723/319/original/8e533fe1a135a541.png
https://files.mastodon.social/media_attachments/files/111/775/134/628/085/206/original/ca4bd9a7cf782149.png

2024-01-18 01:19:27 GMT

Does anyone know what, specifically, is being referred to in this comment about how Facebook claimed they has too much data that's spread out too much to be able to return an inventory for a particular person?
I could imagine a response like this for GDPR data requests or something, but I'm looking for a reference with an actual quote from Facebook on this, not just the generic idea that this is happening.
https://files.mastodon.social/media_attachments/files/111/774/316/947/119/731/original/29da65e68de013ed.png

2024-01-17 07:37:40 GMT

- reply

If you want to compare antibiotics as a class to the internet (which I'm guessing is the intent, since the original statement is meaningless as written), I don't think it's obvious at all, let alone by many orders of magnitude.
If you just look at personally, and ask someone if they could have the internet, but multiple relatives and/or good friends of theirs die because antibiotics don't exist, or they could have antibiotics but no internet, I suspect most would choose antibiotics.

2024-01-17 07:34:37 GMT

- reply

If you think about how the modern internet has enabled conspiracy theories to flourish, and then add to that all of these mysterious deaths, we'd have conspiracy theories on a level that's hard to imagine.
On the original statement, that "the internet is obviously more important than penicillin by many orders of magnitude", I'd agree if taken perfectly literally, but that's like saying antibiotics are more important than ethernet, only obviously true because you'd use substitutes.

2024-01-17 07:28:19 GMT

It's sort of a funny thought experiment, but what would the world look like with the modern internet but without medicine? Infant mortality rate is roughly 50%, life expectancy is something like 20-30 years, with people randomly and mysteriously dying from infections after getting a small cut, etc.
Inequality due to wealth would be reduced because rich people would suffer from all of the same causes of death since this presupposes that no one knows how to treat or prevent any illness.
https://files.mastodon.social/media_attachments/files/111/770/100/415/320/940/original/0ce7b45f12e8e6d7.png

2024-01-16 06:14:47 GMT

I find it mildly interesting that substack has succumbed to user-hostile antipaterns like every other major blogging platform before it (medium, blogger, etc.).
Early in substack's life, one of the pitches was that, by having users pay for content directly (as opposed to via ads), they would be incentive aligned with users and wouldn't need to do the bad stuff other sites to, but a lot of the bad stuff falls out of engagement-optimized ranked feeds, not ads!
(4th screenshot is link from 3rd)
https://files.mastodon.social/media_attachments/files/111/764/147/938/763/452/original/f50cc901ad250d7c.png
https://files.mastodon.social/media_attachments/files/111/764/148/263/892/281/original/f77a2705a0b74b1b.png
https://files.mastodon.social/media_attachments/files/111/764/148/625/027/802/original/1085e1361eb80682.png
https://files.mastodon.social/media_attachments/files/111/764/149/508/246/436/original/e37815ed6ac11f71.png

2024-01-15 07:53:29 GMT

How good as ChatGPT at solving Advent of Code?
https://www.themotte.org/post/797/chatgpt-vs-advent-of-code
Not very good, it turns out.
https://files.mastodon.social/media_attachments/files/111/758/879/780/328/903/original/5773a1bf81e320b9.png

2024-01-12 21:06:46 GMT

- reply

@npub1qm9…l6jg As explained in the post, I think that's both wrong and also not what he said.
But let's say he meant that. Why didn't he just say that instead of making a bunch of much stronger (including some obviously incorrect) claims?

2024-01-12 18:36:35 GMT

Against Epistemic Learned Helplessness:
https://www.patreon.com/posts/71292768
A somewhat old Patreon post, but a bit relevant now since the piece it's arguing against is #1 on HN.

2024-01-11 09:31:55 GMT

- reply

IME, Gmail's false positive and false negative rate has been worse than every random uni dept email address I had as well as when I had an email address on a server with no spam filtering and I did it client side. With Gmail, I get https://mastodon.social/@danluu/109919293479394809, https://twitter.com/danluu/status/1514019521123663873, etc., and I've never had it nearly that bad with small hosts.
Is this a situation like https://mastodon.social/@danluu/111672597118426587, where people just repeat something without ever checking it?

2024-01-11 09:22:15 GMT

Do you run your own mail server and, if so, how easy or hard is to keep on top of spam?
I've heard that it's impossible because of all the spam that you get, but I have an old uni dept email where they only recently migrated off of their own server and I never had an issue with spam when they were running the show.
And likewise for when I ran my own spam filters on my email client on a server without spam filtering.
How are these naive solutions so good if this problem is impossible?

2024-01-10 02:34:45 GMT

- reply

One could argue that I should optimize my process and only do a single pass, but this two-pass process is done instead of the various standard techniques people suggest, like outlining or w/e.
When I've compared notes to people who do standard stuff, my process is much faster, so I don't know of standard advice that would help.
I think just doing a single pass without using some kind of organizational technique wouldn't work unless my working memory was like 100x larger, so that's also out.

2024-01-10 02:31:11 GMT

It's an extreme minority opinion among programmers that typing speed matters, but I started writing again recently and I feel like the argument in https://danluu.com/productivity-velocity/ is still obviously true.
wc has the source for my last post at 20k words. At a hypothetical 100wpm, that's 3h20m of typing time. I often do two passes and totally re-write it, so it's really 7h40m.
If I were to type at 50wpm, that would add a huge amount of writing time, both in absolute and %-age terms.

2024-01-08 20:14:27 GMT

I feel like this example really gets at what I like about reading Andrew Gelman's blog. Re-weighing yourself 46 times because your bathroom scale is high variance is such an obviously unreasonable thing to do that almost no one would do it, but it's a great example of the kind of process I love to see:
http://www.stat.columbia.edu/~gelman/research/published/standarderror.pdf
https://files.mastodon.social/media_attachments/files/111/722/157/715/657/068/original/f6b8812132c448d6.png

2024-01-04 19:37:49 GMT

- reply

@npub1tkr…x9sz I was referring to the senior -> senior staff promo rate, not including the 1 year thing. But also, I don't think that's true. Big wins were things like finding a O(n^2) array resize code that should've been O(n) (a standard interview question), reading a book about the JVM (which I knew nothing about) and then GC tuning, writing ETL jobs to save data that was getting thrown away after two weeks, etc.
Realistically, almost any programmer could do almost all the things I did.

2024-01-04 19:09:08 GMT

- reply

If you wanted to consistently generate, say, $200M/yr/yr for the company in revenue or product in a company that size, you'd run into a ton of barriers because you'd need to make very large changes that visibly impact the product, which everyone has an opinion on.
In a boring old cost center, you can often find these super high impact changes that you can ship if you can find one engineer on a team who thinks it's a good idea, no manager/director/VP/Cxx approval needed.

2024-01-04 19:05:48 GMT

- reply

I was hired at "senior" and then got promoted twice in three years, to "senior staff". In modern Twitter, I only know of one other person who was promoted as quickly to senior staff (someone who worked in revenue).
It turns out, if you make a company hundreds of millions a year (with revenue ~$4B-$5B, profit ~$0), this can often be ok for your career. Per https://mastodon.social/@danluu/111265677136391207, cost savings work is undervalued, so you need to create more value but, IMO, it's also easier to create the value.

2024-01-04 19:02:08 GMT

I find it interesting that it's standard advice that, if you value your career, you shouldn't work in a cost center or on maintenance.
This seems fairly obviously untrue? As this commenter points out, SREs can and often do have great careers. They seem to be one of the 2 dissenting comments out of 186 total comments!
I was in a cost center at Twitter and this seemed fine? I was a fairly new programmer (I had one year of professional programming experience and was previously a hardware eng), so
https://files.mastodon.social/media_attachments/files/111/699/220/760/803/900/original/24b9440a1cb996d2.png

2024-01-03 23:27:48 GMT

- reply

@npub1smc…lkwf I agree that they're independent and that there's something interesting about reduced durability in the cases where that's true (not cars!), but the stuff that goes viral tends to be these rants that lump in durability and useful lifetime and then make really wrong comments about the useful lifetime, and I think it's also useful to note that the useful lifetime or useful lifetime per dollar has not gotten worse for a lot (most? almost all?) of these items.

2024-01-03 23:10:47 GMT

- reply

@npub17dk…88tt Yeah, I looked this up for washers and dryers because someone linked to an article about how washers and dryers are so much worse now.
Compared to 1960, it looks like you can get a low end washer or dryer for the same nominal price, i.e., < 1/10th the real price. If you want to get the cheapest model model that Consumer Reports gives an very very high score to, then you're looking at double the nominal price, which is still an 80% drop in the real price.

2024-01-03 22:20:33 GMT

- reply

I think the cost thing is another component people really miss in these rants. Adjust for cost, I do think jeans are less durable, but I don't find this to be true for t-shirts compared to when I was a kid. It's technically true that the modern t-shirt wears out sooner but, inflation adjusted, the t-shirt cost 1/8th as much, so I get way more value for my money now.
Likewise for many other common examples people give, like a lot of large appliances, which are better and much cheaper.

2024-01-03 22:18:07 GMT

- reply

And, BTW, if you're willing to replace moving parts like the HDs, it is technically true that those old laptops would last longer. Modern chips have all sorts of failure modes that a 386 never did, e.g., I had an overlocked CPU fail after 3 years due, maybe due to electromigration, and maybe the CPU would've failed after 15 years without overclocking, whereas the 386 CPU will probably operate for 50+, but the 386 costs 4 times more is was useless in practice after only a few years.

2024-01-03 22:09:35 GMT

- reply

@npub1e7v…flg9 These are great examples of poor articles that are, like my other listed example, the exact opposite of what I want.

2024-01-03 22:07:45 GMT

- reply

or you could've gotten an "ultraportable" 8086/88 for $11k that weighed 7 lbs, had 640kB RAM and a 20MB HD (and most models had no HD).
10 years later, an ultraportable (not a high-end one) would be a 3.4lb 500 MHz Pentium III and you could get faster machines as well.
Today, a ten year old laptop is still totally usable and I actually have a 10 y/o Macbook I keep around that still works fine.
Anyway, I'd be interested in actual examples of what lasts longer and doesn't.

2024-01-03 22:00:10 GMT

- reply

I've been thinking about this since seeing the rants mentioned in https://mastodon.social/@danluu/111666646509250420, where OP, oddly, lists things like cars and laptops as things that don't last as long nowadays.
Laptops are one of the worst possible examples. I once used a laptop from 1990. It was a 20 MHz 286 that weighed 20 lbs and cost $5k (2023 dollars). It was obsolete in 2 years. Any laptop from PC Magazine's 1990 state-of-the-art laptop review would've been similar obsolete, although you could've gotten a 386,
https://files.mastodon.social/media_attachments/files/111/694/260/406/089/290/original/2b9666f4b9f67f14.png
https://files.mastodon.social/media_attachments/files/111/694/260/940/534/085/original/f1a4db17a6a2db93.png

2024-01-03 21:48:31 GMT

What's gotten more and less reliable & durable over time?
I'd be interested in seeing a breakdown of this by category, the way people break down inflation and find that some things have gotten cheaper while other things have gotten more expensive, but I'm not sure that exists?
I think it's clear that some things last longer and are more reliable than they used to be (cars, computers, lightbulbs, etc.) and some things are less durable and reliable (jeans, some small kitchen appliances, etc.)

2024-01-03 02:59:08 GMT

How good are decisions, Elo edition:
https://www.patreon.com/posts/95627075
I haven't been posting Patreon stuff on Mastodon, but people with newsletters generally report that most traffic comes from Twitter/Mastodon/whatever and I've actually been paying rent from so I'm going to try posting Patreon posts here for a while to see how that goes.

2024-01-01 22:21:50 GMT

I haven't read the comments but, in general, I haven't found lobsters comments to be higher quality except that there are fewer.
I find it weird how many people think it's worth reading comments only. Something like 99% of comments are irrelevant or incorrect. We can quibble about the %-age and maybe it's 95% or w/e, but there's no way it's 50% or even 80%.
I do find comments useful to get an idea what people in a certain (very large) bubble are thinking, but only doing that seems low value.
https://files.mastodon.social/media_attachments/files/111/683/008/268/746/612/original/9bad246531599118.png

2024-01-01 21:54:13 GMT

- reply

@npub1tej…0xuq AFAICT, CR is the best thing around, but as CR readership declines (long story, probably not for public consumption at the moment), their reliability data is getting less and less representative in various ways. CR readership is highly tilted towards reliable vehicles, so they don't even have data for a lot of very popular vehicles, whereas classic reliable brands are highly overrepresented.
And then

2023-12-31 02:10:03 GMT

How bad are search results? Let's compare Google, Bing, Marginalia, Kagi, Mwmbl, and ChatGPT:
https://danluu.com/seo-spam/
https://files.mastodon.social/media_attachments/files/111/672/593/009/263/927/original/539d8020895dd664.png

2023-12-30 00:44:20 GMT

- reply

There's too much data for a Mastodon thread, but you can see this in reliability surveys of cars, not just in the data itself, but in what questions people even bother asking on reliability surveys.
In my lifetime, Consumer Reports tracked data on and reviewed how well cars started. Because, you know, it used to be common for new cars to not start. Now, a photo of a new car not starting is so silly that a friend of mine saw me trending on the front page of reddit, twice, for posting a photo.
https://files.mastodon.social/media_attachments/files/111/666/594/841/778/745/original/69285d4dc64ff896.png
https://files.mastodon.social/media_attachments/files/111/666/595/480/953/639/original/60792e8f94df57a3.png

2023-12-30 00:38:36 GMT

I find it interesting that one of the most common replies to https://mastodon.social/@danluu/111643026109666672 is something like "this is why I'd never buy a car with a computer in it" / "that's less than 15 years old", etc., and that a lot of people generally think that cars are now less reliable.
I haven't pulled it up again, but from having skimmed it in the past, I feel like the data for cars getting more reliable is very strong and that cars have been getting more reliable basically since the creation of the car.

2023-12-27 21:16:25 GMT

- reply

@npub1hat…egax Yeah. Same with https://danluu.com/empirical-pl/ and https://danluu.com/futurist-predictions/ as well as all of the "made up" research Andrew Gelman points out on https://statmodeling.stat.columbia.edu/.
People seem to just make stuff up in ways that are trivial to fact check. This appears to work, in terms of collection fame, fortune, citations, etc.
When someone points out errors like these, people generally seem to respond like Berkeley's research misconduct representative did and just pretend the issue isn't there.

2023-12-27 20:59:35 GMT

I find it interesting that Matt Walker's Why We Sleep is still widely cited by folks despite Alexey Guzey's fairly convincing rebuttal, "Matthew Walker's 'Why We Sleep' Is Riddled with Scientific and Factual Errors": https://guzey.com/books/why-we-sleep/
BTW, I'm much more pro "getting a lot of sleep" than Guzey and don't find Guzey's arguments for how you can get by on little sleep convincing (I've read what he has to say and have talked to Guzey about this), but that doesn't make Walker's book reasonable.

2023-12-25 20:49:45 GMT

"Unfortunately, a recent software update was not successful. Your vehicle cannot be driven.
Please call customer support"
https://files.mastodon.social/media_attachments/files/111/643/022/744/786/331/original/44093c9550d675c4.jpeg