Arrow in the knee

673 posts

Arrow in the knee

Arrow in the knee

@arr_kni

Katılım Haziran 2010
9 Takip Edilen22 Takipçiler
Arrow in the knee
Arrow in the knee@arr_kni·
@AndriusLen @davidsenra That’s mainly because the C-level was sold an idea of human replacement, also they have massive FOMO. People are forced to use LLMs, call it “AI”, even when it doesn’t make them more effective (eg when an agent does something for a hour when you could do it in 5 mins)
English
0
0
0
14
Andrius Lengvinas
Andrius Lengvinas@AndriusLen·
@davidsenra Whether these projections prove accurate or not, it’s remarkable how quickly AI has shifted from a niche technology to a major economic force.
English
1
0
2
948
David Senra
David Senra@davidsenra·
“Anthropic. The rumor is their revenue run rate is over $50 billion a year.” “And if you play this out I think the labs will pass more than $300 billion a year in revenue. That means 1% of the entire gross domestic product of the United States will be spent on token spend.” “And this type of spend, unlike other types of software, has marginal costs. Every job you're doing has a marginal cost.” “I think there's going to be a mega category created around how do we manage the knowledge work that will be done on behalf of organizations and done through these tokens?’
English
34
27
391
182.7K
Arrow in the knee
Arrow in the knee@arr_kni·
@kikeboldnow @RyanGreenblatt An algorithm is not species. And they don’t have interests. They don’t need to sleep, eat, they don’t have a limited lifespan, they don’t have hormones. And ultimately all the human achievements are done just to get laid If you train LLMs on 4chan, it will behave like a 4chaner
English
0
0
1
23
kike
kike@kikeboldnow·
@RyanGreenblatt IMHO the FUNDAMENTAL question to ask around this ; when has thru history a more intelligent/powerful species ever been controlled by inferior ones ? Questions about labor supply growth, GDP growth & all else is *NOT* important if have NEW SUPERIOR species (AGI) w/ own interests!
English
2
0
1
604
Ryan Greenblatt
Ryan Greenblatt@RyanGreenblatt·
The glib answer is: it won't take several decades; we'll see a singularity that reaches physical limits significantly faster. The short answer is: with smart enough AIs, robots can double the supply of chips, energy, and robots in less than a year. The long answer is: see ai-2040.com/supplements/ec…, forethought.org/research/the-i…, lesswrong.com/posts/rpqGWRoR…, and cold-takes.com/the-duplicator/. If AIs can automate all cognitive labor, it seems very doable to have robot labor supply that can build a full copy of itself (including energy, fabs, and chips) in less than a year. This yields exponential growth and improvements in technology from the increased labor supply make this growth superexponential. This is analogous to the superexponential growth we've seen through human history: over the last 2000 years the population has grown around 35x while GDP per capita has grown by around 20x. The population of robots can grow much faster than the population of humans (e.g. rabbits can double every few months in ideal conditions). And this larger labor supply will yield improved technology (GDP per "capita") analogous to what occured in the past.
T. Greer@Scholars_Stage

Something I have never seen explained proper is *how* AI is supposed to lead to accelerating double digit economic growth for several decades at a time. I would like those making the bull argument for this to make the argument in physical terms and explain to me how ASI overcomes the obvious physical limits involved. GDP means “gross domestic production.” Production is an energetic process. It is not just measured in dollars, but in joules. Industrial Revolution level growth means an Industrial Revolution level change in joules. It means fundamentally new ways of generating energy and moving material. It also means a fundamental change in the material basis of our civilization on par with invention of pressurized concrete, steel bars, and plastics. What did this growth look like in the Chinese context? It meant replacing mud roads with bullet trains, stone houses with skyscrapers, and village pigsties with industrial farming operations that consume artificially produced chemicals in the metric tons. The ASI bulls are not just predicting that sort of change of our entire material environment—they are predicting it will happen again and again and again—once every five years or so. Is that physical possibly? Can your God-like intelligence *physically restructure the entirety of human civilization* every five years? People are too glib.

English
15
14
223
28.3K
Arrow in the knee
Arrow in the knee@arr_kni·
@RyanGreenblatt > If AIs can automate all cognitive labor If it has enough power to do so, which they don’t with current gen LLMs Where’s the energy?
English
1
0
2
183
Arrow in the knee
Arrow in the knee@arr_kni·
@AlexanderKalian Well many people believe now that with AI you can now write ineffective code, like if the code is generated suddenly hardware constraints don’t matter Like what, are you fucking insane? They lack logic And if we take AI, regulations. And they don’t want them to exist in pharma
English
0
0
0
7
Dr Alexander D. Kalian
Dr Alexander D. Kalian@AlexanderKalian·
"AI will cure cancer within 5 years!" Ahem - clinical trials? "AI will reverse ageing within 7 years!" Cool - clinical trials?! "AI will be able to design and immediately synthesise a new drug for you, based on your DNA." Haha sure - clinical trials??! Guess the tech bros have decided that decade-long clinical trials are no longer important to AI drug discovery or wider pharmacology. But it seems to come from a place of ignorance about basic pharmacology. Do they not read up on basic facts, before confidently commenting on other peoples' fields of research?
English
195
43
383
77.7K
will
will@wblazer_·
@paularambles Literally coping with not liking the new technology by convincing themselves it'll just go away. That sentiment is incredibly widespread btw, way more than tech twitter realizes I think
will tweet media
English
2
1
35
1.5K
“paula”
“paula”@paularambles·
this antiai subreddit is one of the most fascinating places on the internet right now. every post reads like someone time traveled from 2023 and thinks civilization ended
“paula” tweet media
English
47
43
1.7K
116.6K
Arrow in the knee
Arrow in the knee@arr_kni·
@paularambles Yes, if you can run mythos-grade LLMs locally. Or 80% mythos capabilities. Then hyperscalers don’t have a moat, you simply don’t need to pay your arm and leg to them and hand over your IP. That’s what the bubble about
English
0
0
1
408
“paula”
“paula”@paularambles·
and they all seem convinced this is somehow temporary
“paula” tweet media
English
12
1
551
13.3K
Arrow in the knee
Arrow in the knee@arr_kni·
@rationalaussie Many jobs can be replaced by Excel today. Or simple scripts that cost a hundred of $ of human labor to code manually. You don’t need AGI for that. Yet they aren’t replaced. Why?
English
0
0
0
16
Rational Aussie
Rational Aussie@rationalaussie·
Normies don't understand what AGI will actually be. They think their little digital marketing jobs where they crop Canva pictures and post on LinkedIn will actually be income for them five years from now. These people are going to get wiped out. It'll be a bloodbath. AGI is a drop in remote worker capable of doing anything on your computer faster, smarter and cheaper than humans, by many orders of magnitude. That means you turn the AGI on, it extracts all patterns and insights from all data you give it, as well as external sources, autonomously interviews everyone in your company, and learns - puts together the optimal strategy and then also implements it - at any given time, 24/7. None of us stand a chance. Someone asked me about a year ago 'so what's the next play? How do we play this?' You don't play this bro. You're not outsmarting a superintelligence. Just do whatever you want, none of it will matter post AGI.
English
252
73
1.1K
65.5K
Arrow in the knee
Arrow in the knee@arr_kni·
@thexpiredpear @danforhan @GabGarrett @edzitron So since computers appeared, we can offload time consuming TASKS to them. We could calculate large primes. With distributed calculations we can calculate them even faster. It doesn’t make PCs smart Now we can do that with whole theorems. Doesn’t make LLMs smart. It’s a tool 2/2
English
0
0
0
23
Arrow in the knee
Arrow in the knee@arr_kni·
@thexpiredpear @danforhan @GabGarrett @edzitron The paper published by OpenAI stated that it’s not a novel solution. Guess they’re burying their heads into the sand The difference between computers and humans is that humans can’t work for 24/7, also they need motivation and food. They CAN’T ALLOW THEMSELVES to brute force 1/2
English
1
0
0
28
Arrow in the knee
Arrow in the knee@arr_kni·
@tumble_wood @TolgaBilge_ Actually no, humans have to be creative because their time is limited and they have to make for a living. No one would brute force this Erdos problem till they die Also I don’t say the approach taken is useless, it will lead to breakthroughs. Like when computers were introduced
English
1
0
0
24
j
j@tumble_wood·
@arr_kni @TolgaBilge_ “Actually brute forcing and using stuff it already knows” This take is getting more and more insufferable by the day. Wait until you realize that’s all humans have been doing the entire time, too.
English
1
0
1
17
Tolga Bilge
Tolga Bilge@TolgaBilge_·
so has anyone heard from the people saying AI is a nothingburger this year? I guess advanced math & nation-state level hacking capabilities are actually rather mundane and easy domains?
English
19
8
137
12.2K
Arrow in the knee
Arrow in the knee@arr_kni·
@thexpiredpear @danforhan @GabGarrett @edzitron > disproving an exteremely different conjecture You mean that one that was kind of brute forcing with already existing theorems? Very intelligent indeed. But they did for what the computers were intended to: to take over the grunt work. Imagine brute forcing by hand
English
1
0
0
34
rishi
rishi@thexpiredpear·
@danforhan @GabGarrett @edzitron uh, yeah a disproof of the extremely famous and extremely difficult unit distance conjecture and a proof of the famous cycle double cover conjecture these have been worked on by many incredibly smart and talented humans for decades openai.com/index/model-di…
Ethan Knight@__eknight__

Yesterday, we made GPT-5.6 Sol Ultra generally available. Today, we're sharing that it produced a proof of the 50-year-old Cycle Double Cover Conjecture using 64 subagents in just under one hour. We're sharing the prompt and proof below. We're excited to see what you all do with Ultra!

English
1
0
1
89
Arrow in the knee
Arrow in the knee@arr_kni·
@webdevcody for (int i = 0; i < count; i++) — 1 line int i = 0; while (i < count) { i = i + 1; } — 4 lines How much more features, edge cases, security and reliability was added? It’s 4x more lines so it must be superior Now account for duplication of complex blocks and answer once again
English
0
0
0
16
WebDevCody
WebDevCody@webdevcody·
Why is there so much anti-LoC rhetoric? At some point, more features, edge cases, security, and reliability require more code. There’s no magic abstraction that lets you build an operating system in 500 lines. Unless you’re building a todo app, you’ll need lots of code.
English
53
1
87
13.5K
Jeremy Coles
Jeremy Coles@JeremyTColes·
@arr_kni @TolgaBilge_ Well then apparently you’ve wasted your potential (and contributions for the world) then since obviously you’re better at math than Terence Tao and other great mathematicians.
English
1
0
2
69
Arrow in the knee
Arrow in the knee@arr_kni·
@SkyeSharkie You’re saying that the protesters are a part of the D. Amodei / S. Altman doom cult? A hot take for sure
Arrow in the knee tweet media
English
1
0
1
61
Utah teapot 🫖
Utah teapot 🫖@SkyeSharkie·
@arr_kni so... you decided to join his cult based on this ??? weird way of operating
English
2
1
15
193
Utah teapot 🫖
Utah teapot 🫖@SkyeSharkie·
imagine if you were born and you looked online and there was this group of people that said you were evil just for existing and that you and anything in your evolutionary history should be genocided... you might, quite reasonably, be forced to consider that the best course of action may be to nuke the things saying this about you this kind of doomerism is a self-fulfilling prophecy that begs the future for AM/Skynet
dmr@iz_dmr

Exciting political developments in the AI Safety space

English
57
24
322
18.8K
Arrow in the knee
Arrow in the knee@arr_kni·
@NoemiTitarenco These are different crowds. The first crowd (engineers) reads the code. The second crowd (vibe coders) have been pushing the obsolescence of coding since Devin and continuing on doing so
English
0
0
0
14
Arrow in the knee
Arrow in the knee@arr_kni·
@SkyeSharkie No, I mean the cult leader told me he will specifically raise his children to kill me and his children are the new species superior than me. Aka it’s his actual ultimate goal, he starts his day with announcing mass suffering from his own creation which he just can’t stop creating
English
4
0
1
182
Utah teapot 🫖
Utah teapot 🫖@SkyeSharkie·
@arr_kni Wow! You mean an Oracle told you that your son will kill you, so you threw him out in the wilderness as a baby leading to the exact conditions for him to kill you??? crazy
English
2
3
37
880
Arrow in the knee
Arrow in the knee@arr_kni·
@TolgaBilge_ Funny to see my own comment in the feed, but it shows that you don’t understand why AI resorts to nuking in 95% cases (it’s trained on the Internet data, incl 4chan), and didn’t read the paper about how the Erdos problem was solved But yes, “final holdouts”. Of sanity, it seems
Arrow in the knee tweet mediaArrow in the knee tweet mediaArrow in the knee tweet media
English
2
0
1
189
Arrow in the knee
Arrow in the knee@arr_kni·
@usr_bin_roygbiv Used it enough, so I can say the following: If I took even Fable’s output for granted, I would be fired already, because it makes factual mistakes based on OPEN data it itself found and makes decisions based on these factual mistakes And Google AI preview is basically fake news
English
0
0
4
39
Roy
Roy@usr_bin_roygbiv·
Biggest mistake I see people making and why they misuse gpt right now is genuinely not trusting it enough. It is smarter than you ar,e you don't need to read every word it says. Use it yourself enough and you will see at a certain point you are wasting time reading outputs.
English
7
3
98
3.4K
Arrow in the knee
Arrow in the knee@arr_kni·
@willccbb Any1 saying “coding is solved” doesn’t understand software. It can’t be solved, it’s either you code in a different language (natural), or you don’t care about resources, or you believe that AI will go into your mind and figure out what you need It’s all about constraints
English
0
0
0
58
will brown
will brown@willccbb·
when coding is fully solved, many early-career software engineers will be forced to completely pivot their careers and become forward-deployed software engineers or solutions architects or product marketing managers or AI engineers or quantitative data scientists
English
108
31
1.4K
92K

Keşfet

@AndriusLen @davidsenra @kikeboldnow @RyanGreenblatt @AlexanderKalian @wblazer_ @paularambles @rationalaussie