|
haveblue posted:clearly that person was hari seldon’s very distant ancestor Christ I love those books
|
#
?
Apr 6, 2023 02:36
|
|
|
|
| # ? Apr 28, 2024 11:55 |
|
|
ADINSX posted:llms are a good counter point to my argument. Since the last time I’ve thought about this they’ve become the hot new thing. Maybe the kinks can be worked out and it really will be the hot new thing, who knows half the people using them seem to want to work more kinks in, though
|
|
#
?
Apr 6, 2023 02:36
|
|
|
ADINSX posted:llms are a good counter point to my argument. Since the last time I’ve thought about this they’ve become the hot new thing. Maybe the kinks can be worked out and it really will be the hot new thing, who knows even without the "kinks" it's a tool with a limited set of practical applications. it may end up being a stepping stone to something more useful, but idiots trying to use them to drive their cars or write software or whatever aren't really representative of what the tool can or could do.
|
|
#
?
Apr 6, 2023 02:39
|
|
|
infernal machines posted:even without the "kinks" it's a tool with a limited set of practical applications. it may end up being a stepping stone to something more useful, but idiots trying to use them to drive their cars or write software or whatever aren't really representative of what the tool can or could do. Yeah I agree, there is a ton of hype but it doesn’t feel as big though as ubiquitous high speed internet connections in the late nineties and aughts, or ubiquitous smartphones in the 2010s, but time will tell I guess. I’ve also noticed that as these advances get less impactful, the hype machine tries harder and harder to play it up, like it’s life depends on it. I’ve said it before but: no one had to convince me to buy an iPhone, or get cable internet ADINSX fucked around with this message at 02:48 on Apr 6, 2023 |
|
#
?
Apr 6, 2023 02:45
|
|
|
it's going through a hype cycle that will eventually settle down and it will be used for the handful of things it actually can do and everyone else will look very stupid in hindsight, or they would if anyone remembered anything for longer than 10 minutes
|
|
#
?
Apr 6, 2023 02:48
|
|
|
madmatt112 posted:Christ I love those books the tone on forward the foundation was impressively grim
|
|
#
?
Apr 6, 2023 02:50
|
|
|
a new pattern of online communication i'm seeing nearly everywhere lately is people chiming in on discussions to say something to the effect of "i don't know anything about this topic, but look at this wall of bullshit GPT spit out when i asked about it!" it feels like the web has been invaded by thousands of terry davises. a bit less religious and openly racist, but a similar kind of unprompted pseudorandom gibberish
|
|
#
?
Apr 6, 2023 02:50
|
|
|
cool, they automated d&d posting somebody digitized owlofcreamcheese's brain
|
|
#
?
Apr 6, 2023 02:55
|
|
|
Internet Janitor posted:pacific rim is an earnestly straightforward feel-good action movie about overcoming adversity (in the form of both personal animosity and giant alien monsters) with huge robots that are literally powered by friendship and mutual understanding ive never seen it
|
|
#
?
Apr 6, 2023 03:19
|
|
|
infernal machines posted:yeah, i doubt we're seeing the end of history, even if it were caused by our own hubris. because we'll be dead by then?
|
|
#
?
Apr 6, 2023 03:24
|
|
|
if you're lucky
|
|
#
?
Apr 6, 2023 03:26
|
|
|
PIZZA.BAT posted:i'd say the 2010s gave us id say your too deep into the poo poo if you think the c l o u d is a fundamental shift on par with "the pc" or "consumer smartphones"
|
|
#
?
Apr 6, 2023 03:27
|
|
which isn't immediately noticeable to end users but definitely had an impact
|
fart simpson posted:ive never seen it good post lol
|
|
#
?
Apr 6, 2023 03:27
|
|
|
at this point most orgs have clocked that "the cloud" is overpriced bullshit for many use-cases, but it definitely had a longer half-life than nfts or "metaverses" did, and there are at least some situations where it's genuinely beneficial or at least cost-effective. took a solid decade for reality to sink in at the rate they're going i think llms are going to be the same kind of slow burn disaster as cloud computing, with a similar endgame, but in the meantime they're going to wreck a bunch of poo poo as collateral damage Internet Janitor fucked around with this message at 03:44 on Apr 6, 2023 |
|
#
?
Apr 6, 2023 03:42
|
|
|
infernal machines posted:cool, they automated d&d posting I refuse to believe OOCC has been uploaded to the Matrix for the simple reason that it would prove he was right about everything
|
|
#
?
Apr 6, 2023 03:49
|
|
|
madmatt112 posted:good post lol thanks
|
|
#
?
Apr 6, 2023 03:54
|
|
|
Internet Janitor posted:there's probably still a ton to learn about physics, material science, biology, etc, but we may be reaching the point where made up societal factors prevent significant progress in science this is the social end to my thoughts above, which were purposely naively assuming that physical limits would be the ultimate slowing factor, but this feels more likely. also I have a sneaking suspicion that computers are the latest example of low hanging fruit - we got almost all of the polynomial and below solutions and what remains is the slow hard slog of NP-hard problems via heuristics or careful, specific solutions, and that’s (one of the main) reason we slowed down in the 2010s
|
|
#
?
Apr 6, 2023 03:56
|
|
|
fart simpson posted:id say your too deep into the poo poo if you think the c l o u d is a fundamental shift on par with "the pc" or "consumer smartphones" hmm this feels right but only from the obvious consumer point of view. the concept of distributed computing and high availability, even commodified, being deployed widely is a pretty cool development from a “theoretical” point of view. not sure that came out right, does that track?
|
|
#
?
Apr 6, 2023 04:02
|
|
|
every task neural networks are aimed at is deep in np, its just that they have it monte carlo style instead of big dpll sat things where they have it las vegas style
|
|
#
?
Apr 6, 2023 04:06
|
|
|
Enderzero posted:also I have a sneaking suspicion that computers are the latest example of low hanging fruit - we got almost all of the polynomial and below solutions and what remains is the slow hard slog of NP-hard problems via heuristics or careful, specific solutions, and that’s (one of the main) reason we slowed down in the 2010s i agree. current-day machine learning is arguably a last gasp of trying to use brute force to solve problems in a way that "scales" without requiring insight, invention, or discovery. insert data, turn the crank, get a solution, hope it's within a reasonable epsilon of acceptable there are more robust ways to handle many of the things gpt is being thrown at, but they are much more time-consuming and expensive to develop. for example, compare the kinds of painstaking effort it took to curate the factual databases behind wolfram alpha and the openai approach of "scrape reddit, ignore copyright, dump it in a pile, assume truths are in there somewhere". alpha doesn't handle natural language queries as well as chatgpt, but when it interprets a query incorrectly it's obvious, and it doesn't just pull answers out of its rear end
|
|
#
?
Apr 6, 2023 04:07
|
|
|
Internet Janitor posted:i agree. current-day machine learning is arguably a last gasp of trying to use brute force to solve problems in a way that "scales" without requiring insight, invention, or discovery. insert data, turn the crank, get a solution, hope it's within a reasonable epsilon of acceptable this sounds ideal for me as i dont have any insight
|
|
#
?
Apr 6, 2023 04:08
|
|
|
Internet Janitor posted:i agree. current-day machine learning is arguably a last gasp of trying to use brute force to solve problems in a way that "scales" without requiring insight, invention, or discovery. insert data, turn the crank, get a solution, hope it's within a reasonable epsilon of acceptable
|
|
#
?
Apr 6, 2023 04:08
|
|
|
i always thought cloud stuff felt like marketing mumbo jumbo because its not really a tangible thing but now that kinda feels like a convenient excuse instead
|
|
#
?
Apr 6, 2023 04:09
|
|
|
bob dobbs is dead posted:every task neural networks are aimed at is deep in np, its just that they have it monte carlo style instead of big dpll sat things where they have it las vegas style exactly, we’ve got most of the exact solutions, and have moved onto statistical, close enough, “good” solutions, which are almost vibes based - you know how to finesse solutions but are never closed form solving it - and building those leaky abstractions on top of each other is going to have a lot more art in addition to the science. which also means slow going - you don’t get the quantum leap of exact solutions.
|
|
#
?
Apr 6, 2023 04:10
|
|
|
Internet Janitor posted:i agree. current-day machine learning is arguably a last gasp of trying to use brute force to solve problems in a way that "scales" without requiring insight, invention, or discovery. insert data, turn the crank, get a solution, hope it's within a reasonable epsilon of acceptable lol we basically got 90% with 90% of the effort, and the last 10% will take another 90% of the effort
|
|
#
?
Apr 6, 2023 04:11
|
|
|
fart simpson posted:this sounds ideal for me as i dont have any insight this is silly, almost by definition a jolly contrarian like yourself will hit on some serious insights
|
|
#
?
Apr 6, 2023 04:12
|
|
|
oh while I’m here, quad posting.
|
|
#
?
Apr 6, 2023 04:13
|
|
|
i feel like the events of the 20th century were a freak occurrence that broke humanity's brain, but we're starting to realize that we're not on the path to a technological singularity, whereas in the 70s it was just obvious to everybody that there would be people living on the moon and on mars and such by 2015. compared to the technological progress of the 20th century, no, today's 21st century smartphones are not a particularly revolutionary technology and i'm tired of pretending that they were. the 20th century brought us radio (with its beginnings at the tail end of the 19th century), automobiles (boo hiss), aviation, applied nuclear physics, computing, and transistorized integrated circuits. and the common factor in all of these developments was that they were frantic races to make practical realizations of the not-very-well-explored underlying physical principles before some other competitor got there first, everybody knew that these were vast and fertile unexplored fields of technology. we don't have anything like that even on the horizon right now. biotech certainly doesn't have any, and if there are going to be any huge technological breakthroughs in the next 100 years they will probably be biology related. for one reason or another though biotech moves fairly slowly. the ethical constraints around experimenting on living beings instead of silicon ingots or airfoils in wind tunnels are probably a factor. i think, without any concrete evidence to back this up, that humanity goes through alternating periods of technological and social development, and right now a number of social developmental issues are reaching a breaking point. think about how industrialization in the 18th century and its accompanying thirst for imperialist expansion into indo-china and africa opened up a pandora's box of social strife that the broader world is still grappling with for instance. if i had to guess i'd say that the next hundred years are going to be defined by humanity's collective response to climate collapse and an economy built on the fundamental assumption of long-term exponential growth, rather than transformative technological development per se. or maybe things will limp along as they are for several generations, who knows. the byzantine empire lasted for 400 years and i'm sure that 100 years in some people were confidently predicting that it was on its last legs.
|
|
#
?
Apr 6, 2023 04:14
|
|
|
Beeftweeter posted:i always thought cloud stuff felt like marketing mumbo jumbo because its not really a tangible thing but now that kinda feels like a convenient excuse instead its a way to convert revenue into recurring revenue
|
|
#
?
Apr 6, 2023 04:17
|
|
|
Sapozhnik posted:i feel like the events of the 20th century were a freak occurrence that broke humanity's brain, but we're starting to realize that we're not on the path to a technological singularity, whereas in the 70s it was just obvious to everybody that there would be people living on the moon and on mars and such by 2015. compared to the technological progress of the 20th century, no, today's 21st century smartphones are not a particularly revolutionary technology and i'm tired of pretending that they were. 
|
|
#
?
Apr 6, 2023 04:18
|
|
|
Sapozhnik posted:compared to the technological progress of the 20th century, no, today's 21st century smartphones are not a particularly revolutionary technology and i'm tired of pretending that they were. I think you’re right in the sense of revolutionary, but it really was that big of a deal in the evolutionary sense. they are an incredible combination of computing, materials science, human factor design, communication, networking, sensing, and more all meeting at a mature convergence. a capstone, not a rebuilding of everything edit: this is just a small squabble, the rest of the post is a+
|
|
#
?
Apr 6, 2023 04:19
|
|
|
4lokos basilisk posted:yeah what happened to stuff like that microsoft kinesis something 3d spatial sensor that was immediately used in all student robot projects? I was at an experimental noise show seven years ago that used a dancer in front of one of those things to create music. It worked surprisingly well.
|
|
#
?
Apr 6, 2023 04:20
|
|
|
Sapozhnik posted:the ethical constraints around experimenting on living beings instead of silicon ingots or airfoils in wind tunnels are probably a factor. cowards too bad the bitcoiners are too stupid to do effective research, they have the ethical flexibility we need to get results
|
|
#
?
Apr 6, 2023 04:23
|
|
|
mystes posted:What evidence do you have that it's a "last gasp" rather in the process of improving? probably not the best way of phrasing it, admittedly it's not so much that i think this approach will immediately peter out, but it's a kind of an ontological dead-end. llms are the most general and accessible machine learning tools to ever exist, with no objective way of measuring how effective or reliable they are. GPT-3.5 is generally considered better than 3 and worse than 4, but it's all vibes-based. problems are not reproducible, improvements are not falsifiable, and everything depends on irreplaceable datasets of dubious origin
|
|
#
?
Apr 6, 2023 04:23
|
|
|
i wouldn't say there hasn't been any technological development either. it doesn't matter how old you are, if you can read this shits very different today than it was when you were born the pace of development might be slower but that doesn't mean it's ground to a complete halt. i also don't really have a problem with that happening anyway  we can fix or at least mitigate most problems we have today. we just choose not to
|
|
#
?
Apr 6, 2023 04:24
|
|
|
Beeftweeter posted:i wouldn't say there hasn't been any technological development either. it doesn't matter how old you are, if you can read this shits very different today than it was when you were born this is all true but also nobody is arguing against it 
|
|
#
?
Apr 6, 2023 04:26
|
|
|
we cured bein fat but it costs 1400/mo. or 900/mo for the newer version
|
|
#
?
Apr 6, 2023 04:27
|
|
|
bob dobbs is dead posted:we cured bein fat but it costs 1400/mo. or 900/mo for the newer version what happens when you stop taking semaglutide
|
|
#
?
Apr 6, 2023 04:28
|
|
|
bob dobbs is dead posted:we cured bein fat but it costs 1400/mo. or 900/mo for the newer version in a demonstration of the dictum above that we choose not to, it don’t cost 1400. we are charged 1400
|
|
#
?
Apr 6, 2023 04:29
|
|
|
|
| # ? Apr 28, 2024 11:55 |
|
|
you get fat again lol
|
|
#
?
Apr 6, 2023 04:29
|
|