ππ΄π’π―π«@gwernOct 2Pleased at latest dark mode fixes.
How to do images? Inverting photos/color looks crazy. But not-inverting graphs is blinding white, defeating the point.
My solution: ImageMagick detects monochrome images, I manually tag misses, dark mode inverts only those; +original on hover. pic.twitter.com/sGn2xBapWV
17,199
793
4.6%
View Tweet activity
ππ΄π’π―π«@gwernOct 10I keep wanting to fiddle with the metadata...
"The web developer of the future will be a single designer with a smol cat, a medium-sized laptop, and a large neural network: the neural network optimizes for traffic, and the laptop is sat on by the cat to distract the designer."
ππ΄π’π―π«@gwernOct 8For kicks we experimented with github.com/mnater/Hyphenoβ¦ to get better hyphenation in Firefox (and hyphenation at all in Chrome), but it makes page loads bad on IE (something about the ToC) and caused other problems, so that was no good.
13,836
70
0.5%
View Tweet activity
ππ΄π’π―π«@gwernSep 17Every few years I check en.wikipedia.org/wiki/Berserk_(β¦ to see if it's finished & I can read it yet...
I just noticed it began in 1989 (!). So considering the age distribution for shonen manga, _Berserk_ may now be older than most of its total readership.
12,990
418
3.2%
View Tweet activity
ππ΄π’π―π«@gwernSep 18I think this is a useful trick in many domains: everyone reports nominal amounts even when those are largely irrelevant. For example in investing - how often do we really care about the absolute return or %, compared to the return over alternatives like brain-dead stock indexing?
12,210
257
2.1%
View Tweet activity
ππ΄π’π―π«@gwernOct 3Worth mentioning l4rz's continued work on S2 nudes (although it doesn't use Tensorfork resources). His latest trick: as images get bigger, 'high-resolution layers' are responsible for increasingly big details in eyes/faces. So spend the parameters *there*: twitter.com/l4rz/status/13β¦
ππ΄π’π―π«@gwernOct 4The GAN transmutation/hybridizing seems to have a lot of potential. People have tried hard to train various kinds of CycleGAN and other tricks to do anime β real, and it's never really worked too well. Blend some StyleGAN models, though... twitter.com/swapp19902/staβ¦
11,473
75
0.7%
View Tweet activity
ππ΄π’π―π«@gwernOct 6(People don't love the human touch. They do love to read in big lessons to cross-country variance and Poisson clusters which are noise.
Look at videoconferencing. Obviously, we could've been doing all this years ago! Why not? Inertia. From the view of history, a minor delay.)
ππ΄π’π―π«@gwernOct 3Some Ganbooru updates: if you soup up StyleGAN and are patient enough in tweaking, you can get... a lot more coherent anime images that I had believed possible when we abandoned StyleGAN back in February or so. twitter.com/AydaoGMan/statβ¦
6,633
120
1.8%
View Tweet activity
ππ΄π’π―π«@gwernOct 6"We will now briefly pause the Rectification of Names for a jubilee of insolvent obligations: liquidate (office) real estate, liquidate bullshit (academia) jobs, liquidate (non-tech) companies, liquidate (media/government) nomenklatura; purge the rottenness, values will adjust."
6,474
206
3.2%
View Tweet activity
ππ΄π’π―π«@gwernOct 6If nothing else, sitting through the delays of 'obvious' things and watching all the misfires and prototypes would teach us a great deal more tolerance for delays, and much more skepticism of claims like 'country X doesn't do online shopping because they love the human touch'.
6,264
72
1.1%
View Tweet activity
Get your Tweets in front of more people.
Use Tweet Activity to track how your Tweets are doing.
Engagements
Showing 28 days with daily frequency
Engagement rate
2.3%
Oct 12
2.1% engagement rate
Link clicks
3.6K
Oct 12
49 link clicks
On average, you earned 128 link clicks per day
Retweets without comments
162
Oct 12
1 Retweet without comments
On average, you earned 6 Retweets without comments per day