𝔊𝔴𝔢𝔯𝔫@gwernFeb 1Nobody was born caring deeply about better modeling of ImageNet, or ALE, or Wikitext. You want to do something cool, useful, & interesting in DL? Do literally anything you are interested in and you will almost immediately find yourself on virgin ground with an endless TODO list.
14,462
404
2.8%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernFeb 1It seems like every week there's someone on Reddit agonizing over how they can get hundreds of TPUs to compete with OA or DM. (Hint: "TFRC".) But how absurd—to a first approximation, ML/DL has been applied to 𝘯𝘰𝘵𝘩𝘪𝘯𝘨 thus far. We're 𝘵𝘩𝘢𝘵 bottlenecked on coders!
14,083
215
1.5%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernFeb 10Random corona-induced change: my local college 𝘳𝘦𝘧𝘶𝘴𝘦𝘥 to take credit cards. Bursar even printed big novelty check to explain how 'checks' work to confused undergrads every year. I had to pay with cash/order last year.
I check today & oh sure 𝘯𝘰𝘸 they take wire/ACH/CC!
13,974
519
3.7%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernFeb 15This would be a great visualization opportunity. First page: list of surviving works. Second to tenth pages: list of known lost works. Tenth to ~hundredth pages (set by capture-recapture estimate of total corpus): Markov chain gibberish representing lost & unknown works.
13,641
87
0.6%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernFeb 22Web browsers are OK enough OSes, I suppose, but I've realized that what they really need are some good hypertext implementations for reading.
Another helpful feature: double-click on the titlebar to collapse a popup temporarily (and sticky it). pic.twitter.com/juy5yZzpS8
𝔊𝔴𝔢𝔯𝔫@gwernFeb 27Just noticed @JeffDean's comment arxiv.org/abs/1911.05289 : in scaling hypothesis, almost *has* to happen. Bio-computers are near physical limits & it's raw compute that matters. So *of course* AI would be very unimpressive for most history & Moore slow down just as AI wakes up! pic.twitter.com/GVy6UypEih
13,124
1,447
11.0%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernFeb 15One thought: face-swapping/editing has been discussed for censoring protesters' faces. But there must be loads of similar chilling effects. Porn is the most obvious example: very nonlinear opportunity costs for the most attractive people, mediated by face recognition. What else?
𝔊𝔴𝔢𝔯𝔫@gwernFeb 16Advice for startup product makers making things intended for practical real-world use: if you don't know what you should be doing, "always be knurling".
𝔊𝔴𝔢𝔯𝔫@gwernFeb 13Don't be dense. This story was widely discussed, at length, by everyone in NYT (everyone from their tech reporters to the IT staff weighed in on Slack!) & outside, and they knew it would be scrutinized if they opted to publish it, a year later.
𝘛𝘩𝘪𝘴 𝘪𝘴 𝘵𝘩𝘦𝘪𝘳 𝘣𝘦𝘴𝘵.
𝔊𝔴𝔢𝔯𝔫@gwernFeb 26Any Mathjax usage is expensive (can double your css/font/HTML). Quite a price to adjust one price. Turns out 'putting one number above another' isn't *that* hard in CSS, and I replaced the LaTeX with simple HTML spans + a CSS 'supsub' class. Much snappier: gwern.net/Lorem#inflatio…
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128"><rect fill-opacity="0"/></svg>)
