Twitter | Search | |
Tweets
Dan Luu Jul 11
Replying to @patio11 @pushcx
There are some funny pricing distortions because of this, e.g., cars that were cool 20-25 years ago. Some people are buying 90s Supras for >= $50k. You can buy a new '20 Supra for $50k! But the 90s Supra was cool when now-middle-aged people with disposable income were teenagers.
Reply Retweet Like
Dan Luu Jul 11
Replying to @pushcx @patio11
I think it's interesting is how much people underestimated the market clearing price for high-end keyboards. There was a decent amount of money being left on the table for at least a decade? The demand has been there for a while. As would say, "charge more".
Reply Retweet Like
Dan Luu Jul 11
The pricing of high-end mechanical keyboards is pretty wild
Reply Retweet Like
Dan Luu Jul 9
Replying to @benskuhn
(this set of tweets isn't about you, it's just a general comment; you're not making comments that indicate any of this, but a pretty decent fraction of VC Twitter and general "tech thought leader" Twitter are, not to mention some prominent tech-adjacent bloggers)
Reply Retweet Like
Dan Luu Jul 9
Replying to @benskuhn
1. They have never paid serious attention to journalism 2. They're willing to make a v. strong statement on a subject where they have no background without even doing a cursory search 3. If they're anti-cancel culture, they're quite happy to participate to cancel the outgroup.
Reply Retweet Like
Dan Luu Jul 9
Replying to @benskuhn
While I think the NYT publishing Scott's real name would be obviously pointless and bad, it wouldn't even be in the top 1000 worst things the NYT has done during my lifetime People proclaiming that it's an egregious action, totally out of line with normal behavior indicate:
Reply Retweet Like
Dan Luu Jul 9
Replying to @benskuhn
For me the most surprising (i.e.,informative to me) things have been the responses of "tech thought leaders", many of whom have indicated that they basically have never paid any attention to journalism but are willing to make extremely strong/confident statements about journalism
Reply Retweet Like
Dan Luu Jul 7
Replying to @rygorous @apenwarr
Like, if that's your biggest mistake, you didn't make any mistakes. I don't get it. Taking twice as long as necessary on something seems absolutely terrible to me. I mean, I'm not beating myself up over it, but I consider that to be a very serious mistake.
Reply Retweet Like
Dan Luu Jul 7
Replying to @rygorous @apenwarr
I kind of wonder if this is a reason I'm baffled that people care so little about execution efficiency. In a post, I mentioned I screwed something up in a way that doubled the implementation effort for a project. Most people thought that was a kind of humblebrag, a fake mistake.
Reply Retweet Like
Dan Luu Jul 7
Replying to @rygorous @apenwarr
I don't think it's a coincidence that we didn't go bankrupt and our competitors did; we had < 100 total headcount, many competitors were at 300+ and we weren't in an industry where you could ignore costs and rely on hypergrowth revenue solving your problems.
Reply Retweet Like
Dan Luu Jul 7
Replying to @rygorous @apenwarr
Most of the QA headcount at competitors went to unit testing, but we only did that for the few areas with so much complexity we thought we couldn't cover them otherwise (stuff like the memory hierarchy, which fundamentally has much of the complexity of a distributed DB).
Reply Retweet Like
Dan Luu Jul 7
Replying to @rygorous @apenwarr
Yeah, agree on integration tests. I think one reason we did relatively well (in terms of effort/headcount) compared to competitors producing similar chips (now all bankrupt; AMD and Intel weren't really competing directly) was a focus on integration tests.
Reply Retweet Like
Dan Luu Jul 7
I don't think any of these differences were significant differentiators between Google and MS in terms of effectiveness, but people seem to want to copy the most prestigious company around so the whole industry copied Google. Even MS, which everyone used to copy!
Reply Retweet Like
Dan Luu Jul 7
I think it's funny how much of this appears to be totally arbitrary, e.g., MS was relatively big on QA/testing as a discipline, but when Google became the company to copy, MS switched over (changed interview questions, remodeled offices to open plan, laid off most testers, etc.)
Reply Retweet Like
Dan Luu Jul 7
IMO a lot of it is this thing you've mentioned a couple times, that testing is a skill and QA is untrendy. One big difference was that QA (not called QA, but the equivalent job) was as respected as development, on the same payscale, with the same career advancement opportunities
Reply Retweet Like
Dan Luu Jul 7
(not including classes of security bug that hadn't been invented yet) 2:1 would be overkill for most software, but 1:10 would go a long way if applied efficiently. In terms of time spent, I think we're already > 1:10 on correctness vs. development, it's just spent inefficiently.
Reply Retweet Like
Dan Luu Jul 7
At the chip company I worked for, we had ~2:1 QA:dev and almost all devs did 0 testing. In 8 years, we had maybe 2 logical bugs that would be "SEV1" at a software company, on a product with more stateful complexity than anything I've touched that's not a distributed database.
Reply Retweet Like
Dan Luu Jul 7
IMO, part of the problem is that people overestimate difficulty, e.g., the OP we're discussing makes the scaling argument that testing is O(2^n) hard. It may be exponentially or combinatorially hard in a sense, but like with many hard problems, simple heuristics can go a long way
Reply Retweet Like
Dan Luu Jul 7
it would've been less work to just think about the edge cases in the first place and then write the test ensuring that class of bug doesn't occur. Obv. you're not going to catch everything, but I think you can reduce total effort pretty significantly.
Reply Retweet Like
Dan Luu Jul 7
But I think that, for a given level of reliability, we often get there in a really efficient fashion, e.g., for the bug in the video, I'd be totally ok with that being marked WONTFIX, but given that it wasn't WONTFIX (it was fixed shortly after the tweet),
Reply Retweet Like