𝔊𝔴𝔢𝔯𝔫@gwernMar 7The real irony here being that our very first WP popups implementation, 3 years & 6 implementations ago, 𝘸𝘢𝘴 dynamic! We didn't have recursive popups or know the WP API we ought to've used ('Mobile'), but nevertheless—𝘁𝗶𝗺𝗲 𝗶𝘀 𝗮 𝗰𝗶𝗿𝗰𝗹𝗲.
𝔊𝔴𝔢𝔯𝔫@gwernMar 14My popups are opinionated. A controversial choice: rewriting all uses of the word 'significant/significance'. Disgusting corruption of language.
Here's a not-even-that-bad example: no less than 3 totally different uses of it—and impossible which to know which w/o reading PDF! pic.twitter.com/hPD4LPnYfK
𝔊𝔴𝔢𝔯𝔫@gwernMar 10So many nice tweaks we can make. The thumbnail is floated right rather than inline, disambigs are deleted as cruft, and we've popped up a separate section link showing the whole section (and note the '§' usage in the titlebar!). pic.twitter.com/4ot9Q4fOPK
16,220
314
1.9%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernMar 17twitter.com/arankomatsuzak… appears to remove the adversarial loss from D's training entirely, in favor of contrastive learning. Works quite well. Now, how to do G...
14,038
38
0.3%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernMar 23What's the absolute simplest concrete way to explain and 𝘤𝘰𝘮𝘱𝘶𝘵𝘦 Bayesian updates even a middleschooler could understand? Laplace, VI, or MCMC won't work.
My suggestion: 'sets of possible worlds' as the semantics, then exact ABC (rejection sampling) to compute updates.
𝔊𝔴𝔢𝔯𝔫@gwernMar 14The amazing asymmetry of computer hardware: in the time it takes to cp the PDFs off the DVD, I can install jtr-jumbo & hashcat-nvidia, read the docs, and get through >170 billion hashes trying to bruteforce the passwords.
𝔊𝔴𝔢𝔯𝔫@gwernMar 21"Maybe good things, like health, intelligence, non-disability, knowledge, or longevity, are... 𝘨𝘰𝘰𝘥?"
[ETHICISTS STAND UP ANGRILY]
"& stereotypes are accurate, common sense useful, and even—2+2=4?"
[PANDEMONIUM. MOB SWARMS PODIUM WITH BURNING TORCHES & PITCHFORKS, SOMEHOW]
10,384
165
1.6%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernMar 7You don't appreciate how densely linked and 𝘣𝘪𝘨 Wikipedia is until you're a week-and-a-half into crawling WP and find yourself cycle detection in directed cyclic graphs & realizing you need Dijkstra's algorithm to do annotations right. Should've used JS for dynamic popups...
𝔊𝔴𝔢𝔯𝔫@gwernMar 21Misc Qs:
1. is mT5 (still) the largest dense NN model publicly available? reddit.com/r/mlscaling/co…
2. Whatever happened to 'superconvergence'?
3. Is there any JS tool for heuristically deciding to invert photos in dark mode? (eg avoiding faces)
9,412
213
2.3%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernMar 6Fixed alpha thresholds are a garment cut to fit everyone and so suit no one: they are far too stringent for the decisions we must make every instant, yet far too spineless for creating sound knowledge.
(Yes, this is about 2020–2021 too.)
9,367
89
1.0%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernMar 7I wonder if any double-entry accounting systems or software try to do inflation-adjustment by default to work in real amounts instead of nominal? Seems like you could handle it by treating as different currencies with exchange rates & at end of year, 1 'transaction' to swap cash.
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128"><rect fill-opacity="0"/></svg>)
