𝔊𝔴𝔢𝔯𝔫@gwernOct 9(Not to criticize Norvig! I think we were just epistemically unlucky. Like heliocentrism, the rational thing was to disbelieve. "OK bro, it's a nice idea, but we clearly do *not* see any parallax, and for that to be true, the universe would have to be... absurdly big numbers.")
7,828
274
3.5%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 30Generative artists sometimes deprecate larger models, saying "glitches are what make it good". But glitches are far more interesting against high-quality.
"An elephant has 5 legs". Lol? I'd rather read a monograph rigorously working out consequences (w/memes mocking 'squares')!
6,722
209
3.1%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 1That moment when you realize your website will never be as nicely typeset as am ASCII game FAQ from 2000 because you don't care enough to git gud. pic.twitter.com/0iVMibsL5i
6,670
1,171
17.6%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 18Design by compression: pg evaluates languages by lines to a web server; Wirth required Oberon compiler features to speed up self-compilation; STEPS remained within 20kloc; some researchers/devs use package repos (eg CRAN or Debian). What different quantities do these minimize?
5,364
143
2.7%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 12I feel like a lot of things could use that trick. Like web serials could have one per major character. I bet authors could commission a set of 5-10 nice decorative designs for $500 easy and it'd make chapters that much nicer to read through. Could also show character progression.
𝔊𝔴𝔢𝔯𝔫@gwernOct 19Worth mentioning is Anima Anandkumar (Nvidia/Caltech); when she posted her enemies list in Dec 2020 to coordinate attacks on Pedros Domingos and anyone associated, people were reporting themselves added for apparently as little as a single Like.
4,965
208
4.2%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 3I know I'm very comforted when I see only 2 researchers on every paper, with no recent papers, one of which owns a company selling virus tests, and a monocasual explanation wildly at odds with minor things like obesity GWASes, w/differences like '50 pounds' vs '3 pounds'...
4,651
140
3.0%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 24Shower thought: bananas are good/cheap but hard to bulk buy because of the ripening problem - can't get a good sequence. Ethylene gas rises, so it can slow a batch by ventilation. But what if we stack bananas between vertical glass plates ||? Top ripens fastest, overall gradient.
4,437
172
3.9%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 9I disagree. People who talk openly about AGI get criticized and mocked a lot more for 'worrying about Terminator' or 'overpopulation on Mars' or 'distracting from the *real* AI bias problems'. Look at Schmidhuber, Sam Altman, Sutskever... "Serious" "respectable" people don't AGI.
4,034
157
3.9%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 12Because it didn't do either one of those? The $1b investment was for equity in the OA LP itself, and the GPT-3 exclusive deal made later on was for access to the code, not the trained model OA runs.
3,676
59
1.6%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 27Calling it now: uranium antique glassware & camera lenses is the new tungsten NFT.
3,610
67
1.9%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 28"Oh, I'm sorry, we can't pay you in real cash. Just in exposure (to 0.01% of the upside). Wait, wait, where are you going?"
3,585
44
1.2%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 5"Root-cause analysis results: powerstrip had exposed power switch; no UPS was available to take over in any power failure; inadequate persistence to disk of files.
Recommended remediations: powerstrip with cover; install UPS; decrease editor timeout; schedule orphanage visit."
3,307
78
2.4%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 15No, "death of the author". I've read too many interviews with creators to believe that their personal preferences or opinions have much correlation with what I or everyone else values/enjoys in their work. The first time, shocking. 100th time, like being nude in front of a dog.
3,186
62
1.9%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernOct 17The Netflix one is better than their current one, which is just the word 'Netflix' in red. (It doesn't even have the fun outline/dropshadows it used to have, blandifying it further.)
3,129
36
1.2%
View Tweet activity
Engagements
Showing 31 days with daily frequency
Engagement rate
3.9%
Oct 31
4.0% engagement rate
Link clicks
1.7K
Oct 31
170 link clicks
On average, you earned 54 link clicks per day
Retweets without comments
0
Oct 31
0 Retweets without comments
On average, you earned 0 Retweets without comments per day