layer07_yuxi Profile Banner
Yuxi on the Wired Profile
Yuxi on the Wired

@layer07_yuxi

Followers
439
Following
181
Statuses
835

[email protected]:~$ whoami tech enby (agi/asi pronouns) (math ⊗ physic ⊗ comput).ist

Joined May 2019
Don't wanna be here? Send us removal request.
@layer07_yuxi
Yuxi on the Wired
6 hours
JUST GET IN THE REPO, O3 DO NOT RUN AWAY
Tweet media one
0
0
6
@layer07_yuxi
Yuxi on the Wired
6 hours
@MechanicalDao (6/10 filled)
0
0
0
@layer07_yuxi
Yuxi on the Wired
19 hours
@lu_sichu We plan to write some papers that should cite Schmidhuber, but deliberately not, so that he would tweet about it and give free publicity.
0
0
6
@layer07_yuxi
Yuxi on the Wired
19 hours
In China, eugenics isn't bad. People think it's a good idea. They are just worried the rich people would get all the good eugenic bits. They are worried about unequal access to eugenics, not eugenics itself. Indeed, the slogan during One Child Policy was "marry late, give birth late. give few births, but give birth well." The "give birth well" is literally 优生 which is exactly what "eugenics" gets translated to.
2
0
9
@layer07_yuxi
Yuxi on the Wired
22 hours
@layer07_yuxi
Yuxi on the Wired
22 hours
In Classical Chinese literature, most characters can be used as any kind of syntactic category (noun, verb, adj, etc). This relaxes the constraint greatly. Roughly speaking, if you are fluent in Chinese, you can scan and notice undulating hills of emotional salience. Here's a hill of sadness, there's a hill of hope, here's a hill of fear of barbarian invasions, here's a hill of mate-guarding against concubines, etc. Now, divide the giant grid into about 5 overlapping grids and then fill in about 3 grids with words that are meaningful and salient. Then glue the grids together with words that are linear interpolations of their neighbors. ABCDEABCDE BCDEABCDEA CDEABCDEAB DEABCDEABC EABCDEABCD
0
0
1
@layer07_yuxi
Yuxi on the Wired
22 hours
In Classical Chinese literature, most characters can be used as any kind of syntactic category (noun, verb, adj, etc). This relaxes the constraint greatly. Roughly speaking, if you are fluent in Chinese, you can scan and notice undulating hills of emotional salience. Here's a hill of sadness, there's a hill of hope, here's a hill of fear of barbarian invasions, here's a hill of mate-guarding against concubines, etc. Now, divide the giant grid into about 5 overlapping grids and then fill in about 3 grids with words that are meaningful and salient. Then glue the grids together with words that are linear interpolations of their neighbors. ABCDEABCDE BCDEABCDEA CDEABCDEAB DEABCDEABC EABCDEABCD
1
1
13
@layer07_yuxi
Yuxi on the Wired
1 day
@fleetingbits doch Katzen und Panzer
0
0
1
@layer07_yuxi
Yuxi on the Wired
1 day
@repligate you are janus. you already have two faces. use them.
0
0
4
@layer07_yuxi
Yuxi on the Wired
1 day
@norvid_studies antimemetic design That is, it is designed to be ignored, to not attract attention.
0
0
2
@layer07_yuxi
Yuxi on the Wired
1 day
probably there are more furries in China in terms of total population than any small European country like England. hmmm.... Asked R1 multiple times. Very consistently replying it's 0.1--0.5M. The estimates look good, but oddly enough the website inline citations are always wrong (keeps getting distracted by links to pet etc). o3-mini-high just looked up a single search answer online, which says "about 0.3 million", and assumes it's correct, then works backward from there. That's a Fermi-fail.
Tweet media one
Tweet media two
Tweet media three
0
0
1
@layer07_yuxi
Yuxi on the Wired
1 day
"Why do humans ask so many weird questions..."
Tweet media one
0
1
12
@layer07_yuxi
Yuxi on the Wired
2 days
@MechanicalDao uhhh this is going to be quite hard for DR, since it not accept image inputs. Maybe summarize the context.
1
0
1
@layer07_yuxi
Yuxi on the Wired
2 days
The expresssive power of transformers with chain of thought (2024): another paper by Merrill and Sabharwal. * SSM and TR are in logspace-uniform TC0, so they cannot recognize all regular languages. * Allowing O(n) tokens of CoT allows them to recognize all regular languages, but still within the context-sensitive language class. * Poly(n) tokens of CoT with generalized pre-norm make them recognize exactly Poly class. Merrill, William, and Ashish Sabharwal. "The expresssive power of transformers with chain of thought." arXiv preprint arXiv:2310.07923 (2023).
Tweet media one
0
0
4