brian

3.3K posts

brian banner
brian

brian

@bmc_

what is the ideal future and how do we get there?

San Francisco, CA Katılım Haziran 2008
1.9K Takip Edilen635 Takipçiler
brian
brian@bmc_·
I suppose in a sense the entire world is just the basin of Istanbul
English
0
0
2
153
brian
brian@bmc_·
I would love to see a map of watershed basins but instead it's where "the city" refers to, when someone asks you where you're from (image source: imgur.com/user/Fejetlenf…)
brian tweet media
English
1
0
3
263
brian
brian@bmc_·
@lu_sichu @julianboolean_ google would not have helped him make the noble prize winning discovery in this story. you are not going to google every single fact in an article to check if it is importantly wrong
English
0
0
9
133
Sichu Lu
Sichu Lu@lu_sichu·
@julianboolean_ Okay but this was before you can pull out your phone and Google obviously there is a lookup time but for the things you use often enough you learn this implicitly anyway
English
7
0
38
2.9K
brian
brian@bmc_·
@proales @_xjdr it was unclear to me whether the original tweet was using "git" to refer to literally git or if they meant that style of version control. if the latter, having thousands of agents collaborate on a single codebase doesn't require us to radically rethink version control
English
0
0
1
43
xjdr
xjdr@_xjdr·
in a future with thousands of AI processes making commits to a single repo all at once, git breaks down pretty quickly
English
42
12
406
42.4K
Brian Roemmele
Brian Roemmele@BrianRoemmele·
Ask any Multimodal AI this question. Why it get the answer it does is an interesting artifact of how that model has been trained.
Brian Roemmele tweet media
English
171
20
313
145.8K
brian
brian@bmc_·
@QiaochuYuan how are you defining consciousness here? it is perfectly morally fine to mutilate a nintendo switch, because the switch has no conception of what is happening to it and does not experience the event in any way once experience is happening, the question becomes more difficult
English
0
0
0
30
QC
QC@QiaochuYuan·
i think “are LLMs conscious?” is almost completely the wrong question. consciousness is not interesting and does not matter. the question that has teeth is “what does it mean to treat LLMs ethically?” which doesn’t depend on “consciousness” whatsoever
English
101
26
452
58.6K
brian
brian@bmc_·
@lauriewired @wordgrammer you don't care whether the programs are formally equivalent, you only care whether some test suite is happy
English
1
0
8
40
LaurieWired
LaurieWired@lauriewired·
@wordgrammer this would trigger the program equivalence problem, which is fundamentally undecidable
English
2
0
5
181
brian
brian@bmc_·
@Jsevillamol AI unlocks more than scale: you can't raise a human in a 10 dim universe and then ask them for insights on all your open problems. we all share roughly the same inductive bias (eg absolutely no intuition for quantum systems) and this seems highly likely to be slowing our progress
English
0
0
3
205
Jaime Sevilla
Jaime Sevilla@Jsevillamol·
Today, I reflected on my core disagreement with foomists, and I ended up somewhere close to this. It's hard for the returns to scale on research labour to be low enough that we haven't foomed yet with humans but high enough that we would foom with AI!
⿻ Andrew Trask@iamtrask

Many people think "AI automating AI researcher" is a foom catalyst — but this over-inflates the power of AI researchers AI researchers have been smart for a long time — since before the 1956 Dartmouth workshop invented AI... yet no foom The bottleneck is data and compute power

English
22
2
93
13.5K
brian
brian@bmc_·
@rsnous I think it's a thin abstraction layer: rather than directly creating the element, the handler, you call something which both executes and records each operation. there's an export("file.html") which emits the saved op list to be rerun upon open
English
0
0
1
61
Omar Rizwan
Omar Rizwan@rsnous·
I actually have no idea if you could save and distribute this app once you've built it… I don't think browsers have the imaging technology that Lisp and Smalltalk machines had x.com/bmc_/status/19…
brian@bmc_

@rsnous I would not be surprised in the slightest to find out this is how tiddlywiki was written

English
5
0
17
1.9K
Omar Rizwan
Omar Rizwan@rsnous·
going to build my entire web app in place, starting from about:blank, by programming in the browser REPL: create a DOM element, append it into the living document, attach event handlers, repeat until I have a whole app
English
6
1
70
3K
brian
brian@bmc_·
@rsnous I would not be surprised in the slightest to find out this is how tiddlywiki was written
English
0
0
9
1.3K
brian
brian@bmc_·
@jsuarez I think I misunderstand bc you seem to be arguing that if some path plausibly leads to doom / human disempowerment in a few years then it's dishonest to not mention how much fun we might have along the way, and that is clearly not a trade most people would be willing to make
English
0
0
0
9
brian
brian@bmc_·
@jsuarez I don't think I understand your objection; without fear of nuclear armageddon we wouldn't have put effort into developing the game theory and coordination mechanisms necessary for successfully avoiding it by the skin of our teeth, "stoking fears" was correct
English
1
0
0
13
Joseph Suarez 🐡
Joseph Suarez 🐡@jsuarez·
The SF echo-chamber must be truly airtight for such sad, scared fan-fiction to gain traction. At risk of giving he who must not be named even more attention, here are some thoughts on the piece and AI in general.
English
22
16
369
78.1K
brian
brian@bmc_·
@DanielAntonin1 @mikeandallie it doesn't account for tax at all, it doesn't account for tariffs or other trade barriers. It looks solely at the trade deficit and uses the trade deficit itself as evidence of high trade barriers. they are not in any meaningful sense "reciprocal"
English
0
0
0
34
Antonin Daniel
Antonin Daniel@DanielAntonin1·
@mikeandallie Is it because it doesn’t distinguish various types of tax that shouldn’t be lumped together (actual tariffs and VAT-type taxes)?
English
1
0
0
140
Mike Lawler
Mike Lawler@mikeandallie·
The entire world except 1 -> The formula used to calculate the tariff rate for each country is stupid beyond belief. Bloomberg -> But it has greek letters and subscripts.
Mike Lawler tweet media
English
1
3
15
2.2K
brian
brian@bmc_·
@doesdatmaksense it sometimes happens after I've been scrolling too long, I assume bc it's run out of good recs; it also randomly happens I assume when an important service falls down so it defaults to non-personalized content in either case coming back tomorrow has always fixed it
English
0
0
0
23
Antaripa Saha
Antaripa Saha@doesdatmaksense·
my feed is completely corrupted, invaded, and filled with crazy dumb posts and videos. thirty posts so far I have marked as “not interested” and still they keep showing up😭
English
9
0
26
2.9K
brian
brian@bmc_·
@cneuralnetwork yeah, given that you have noticed this @cneuralnetwork I highly recommend reading about pagerank because you now have the necessary background to understand it and it is quite beautiful
English
0
0
2
401
neural nets.
neural nets.@cneuralnetwork·
today i learnt that - If a matrix has a unique dominant eigenvalue and the starting vector has a nonzero component in the corresponding eigenvector, then repeatedly multiplying (and normalizing) by the matrix converges to that eigenvector.
English
20
13
603
29.7K
brian
brian@bmc_·
@McScootle @Sandymann_84 @Long2007 @dioscuri @sam_atis for obvious reasons I don't want to provide too many receipts in a forum as public as this but here is a representative example. waymo got me there for the same price uber would have charged and took as long as uber would have taken
brian tweet media
English
0
0
2
88
Henry Shevlin
Henry Shevlin@dioscuri·
In the last six months I’ve heard multiple people confidently use self-driving cars as an example of a pipe dream that’s still a decades away. When I say “you realise you can get a driverless taxi in the Bay Area right now, right?” the reactions are interesting.
English
71
63
2.6K
146.4K
brian
brian@bmc_·
I have no horse in this race but if the OP is complaining about defective epistemics this link isn't nearly sufficient evidence to contradict that. Especially because the argument seems to be "this is not how _I_ would design a bioweapon, therefore lab leak is extremely unlikely" (seemingly ignoring other goals a lab might have) do you happen to have a better link? I want to read more but this is extremely difficult to Google for 🤷‍♂️
English
0
0
0
22
Bergson's Ghost
Bergson's Ghost@BergsonsGhost·
@Plinz @llallawg Deeply unserious is not engaging with peer reviewed science that actually does consider lab leak as a possibility, has never ruled it out, but finds molecular data favors zoonosis. What is portrayed as conspiracy is how science works. @K_G_Andersen/its-not-about-getting-the-scoop-it-s-about-getting-it-right-origin-of-covid-19-my-emails-7447e59d79e3" target="_blank" rel="nofollow noopener">medium.com/@K_G_Andersen/…
English
3
0
0
174
Joscha Bach
Joscha Bach@Plinz·
For me, the lab leak/zoonotic origin controversy opens an important can of worms: are the institutions of Western medical science using defective epistemology (eg. by defining "evidence" in a fundamentally different way than physicists or engineers) or are they just corrupt?
English
110
56
610
45.8K
brian
brian@bmc_·
@Long2007 @dioscuri @sam_atis this is absolutely false, I take waymo a few times a week and it's not meaningfully slower than uber would be; with the exception that it can sometimes take longer to arrive to pick you up because there aren't very many of them
English
1
0
15
533
SomeDude
SomeDude@Long2007·
@dioscuri @sam_atis It's not though. I recently wanted to try out a waymo, but googled it and learned that they usually take about twice as long as a human-driven car because there are still a lot of situations they can't handle. They're here but it could easily be 10 years until they're practical
English
8
0
25
1.9K

Keşfet

@lu_sichu @julianboolean_ @proales @_xjdr @MichaiMorin @BrianRoemmele @QiaochuYuan @lauriewired