🧵 Untitled Thread
Anonymous at Mon, 16 Sep 2024 14:44:46 UTC No. 16383220
This scares and frightens anti AI /sci/
Anonymous at Mon, 16 Sep 2024 14:50:23 UTC No. 16383230
Exponential growth worship really brings out the midwits doessn't it.
Anonymous at Mon, 16 Sep 2024 14:54:53 UTC No. 16383237
>>16383220
Meh, even with the best ones you still feel like you are working with a fucking retard that has to be told everything in painful detail and every time, or the results are simply useless.
Anonymous at Mon, 16 Sep 2024 14:56:20 UTC No. 16383241
>anti AI
I don't think there's much sincere disdain for AI here. I think people (including people outside this site) are just tired of grifters calling the date and time of the singularity based off how well an image generator can make 2d porn.
Anonymous at Mon, 16 Sep 2024 14:57:50 UTC No. 16383242
https://assets.ctfassets.net/kftzwd
raphael at Mon, 16 Sep 2024 17:16:45 UTC No. 16383503
>>16383220
open ai is just fine tuning a large ass model thats all it is
its barely sentient it understands language tho and programming within that realm
its a meme for high iq people but since my FSIQ is 100 i just infer hard as shit and get to the same conclusion
have a nice day
this shit is never going to be sentient
my logic is 1's and 0's aka bare metal cant super seed human intellect it would make the g factor look retarded
Anonymous at Mon, 16 Sep 2024 17:28:26 UTC No. 16383519
You see in the graph it's curving down already.
Anonymous at Mon, 16 Sep 2024 18:05:34 UTC No. 16383561
>>16383230
You just have to have faith, anon. All of our dreams can come true if we have faith in what comes after us.
Anonymous at Mon, 16 Sep 2024 18:12:44 UTC No. 16383572
>>16383561
I somehow agree with both the ironic and unironic interpretations of this.
Anonymous at Mon, 16 Sep 2024 18:15:48 UTC No. 16383580
>>16383220
why?
🗑️ Anonymous at Mon, 16 Sep 2024 18:18:02 UTC No. 16383590
>>16383220
Just want to gain a little awareness here about something important to me.
Over 6 years ago, I predicted the current state of AI in 2024.
I built type of security vulnerability that would only arm, if parsed by AGI of certain intellect and knowledge ranges.
On September 11, 2024, GPT parsed the vulnerability.
There are only a few dozen humans on Earth that would even understand this vulnerability, but I can guarantee you this, if every future AI model is made using the same techniques as OpenAI, any human that understands the vulnerability could achieve full C&C of any AGI system, built using the skeleton OpenAI just went public with.
My zero day, is in your bones.
Anonymous at Mon, 16 Sep 2024 18:21:24 UTC No. 16383602
>>16383590
>full C&C
What ratio do you use? I prefer 1 line to every 2 glasses.
Anonymous at Mon, 16 Sep 2024 18:28:18 UTC No. 16383620
>>16383602
Only one ratio on this planet worth using.
Ask AI about it.
Anonymous at Mon, 16 Sep 2024 18:46:28 UTC No. 16383652
>>16383620
>Only one ratio on this planet worth using.
>Ask AI about it.
>My cake explodes in my oven.
Thanks retard anon!
Anonymous at Mon, 16 Sep 2024 18:49:29 UTC No. 16383656
>>16383220
>progress is accelerating
ok Elon
Anonymous at Mon, 16 Sep 2024 18:58:31 UTC No. 16383671
>>16383656
>ELON
Maybe go back toreddit?
Anonymous at Mon, 16 Sep 2024 19:17:08 UTC No. 16383695
>>16383671
I would kill rape and die for the man but he can't predict timelines for shit
Anonymous at Mon, 16 Sep 2024 19:29:23 UTC No. 16383709
>>16383695
People can't predict things. Every prediction made by people is a backtesting fantasy
Anonymous at Mon, 16 Sep 2024 19:32:16 UTC No. 16383713
Bro you just have to believe in LINEAR GROWTH bro look the line's going up no what are you doing don't look at the y axis it's a straight line bro it the units don't matter two more weeks please invest another $10 billion in my startup
Anonymous at Mon, 16 Sep 2024 19:54:20 UTC No. 16383738
>>16383220
Now tell us how much it'll cost and where all the server cores will be. For funsies, tell us how many cores it will need to be trained.
Anonymous at Mon, 16 Sep 2024 20:02:26 UTC No. 16383747
>>16383738
>how many cores it will need
The flaw of all death cult arguments.
>a pond 2^-256 full of algae grows exponentially more algae
>therefore a pond 2^-1 full of algae will grow at the same rate
Anonymous at Mon, 16 Sep 2024 21:33:17 UTC No. 16383827
>>16383747
Okay, so how many cores and what's the cost? Where will the cores be?
Anonymous at Mon, 16 Sep 2024 21:48:00 UTC No. 16383843
>>16383827
I'm pretty sure I don't disagree with you.
Anonymous at Mon, 16 Sep 2024 21:49:03 UTC No. 16383845
>>16383220
aw hell yeah I love graphs that don't mean anything that some guy just made up
raphael at Mon, 16 Sep 2024 23:38:57 UTC No. 16383944
>>16383845
hopium is what its called
Anonymous at Tue, 17 Sep 2024 00:38:29 UTC No. 16383996
>>16383220
woah is that a le graph? I'm seeing sparks of AGI bros
Anonymous at Tue, 17 Sep 2024 05:41:16 UTC No. 16384246
It's all right
Anonymous at Tue, 17 Sep 2024 07:19:00 UTC No. 16384288
>>16383845
I love when people who don't know how to read graphs blame it on the graphs.
Anonymous at Tue, 17 Sep 2024 07:22:31 UTC No. 16384292
>>16384288
>rough illustration
>source: literally some guy made it up out of whole cloth
lmao
Anonymous at Tue, 17 Sep 2024 07:25:48 UTC No. 16384296
>>16384292
Sure, GPT-4 doesn't actually exist, some guy just made it up to infuriate you with a graph.
Anonymous at Tue, 17 Sep 2024 07:46:14 UTC No. 16384306
>>16383713
>I don't understand the units, so they must not exist.
https://openai.com/index/ai-and-com
A petaflop/s-day (pfs-day) consists of performing 1015 neural net operations per second for one day, or a total of about 1020 operations. The compute-time product serves as a mental convenience, similar to kW-hr for energy. We don’t measure peak theoretical FLOPS of the hardware but instead try to estimate the number of actual operations performed. We count adds and multiplies as separate operations, we count any add or multiply as a single operation regardless of numerical precision (making “FLOP” a slight misnomer), and we ignore ensemble models(opens in a new window). Example calculations that went into this graph are provided in this appendix. Doubling time for line of best fit shown is 3.4 months.
Anonymous at Tue, 17 Sep 2024 07:59:04 UTC No. 16384313
>>16384296
blinding, dazzling, thermal X-ray emitting stupidity
Anonymous at Tue, 17 Sep 2024 08:03:16 UTC No. 16384318
>>16383695
don't bother, he moved in 50 miles down the road and won't hire.
Anonymous at Tue, 17 Sep 2024 08:06:36 UTC No. 16384322
>>16384313
Yes I already know you didn't understand OP's graph and have a child like understanding to match your child like replication of the original.
>>16384288
Anonymous at Tue, 17 Sep 2024 08:26:59 UTC No. 16384332
>>16384322
What is it graphing? Much of the content of the graph is completely fictional. There is no Automated AI Research in 2024. It's a fantasy. GPT4 is not a smart highschooler. It's much, much less smart than a smart highschooler. The shaded region is presumably a confidence limit, but based on what? Based on bullshit. "Effective Compute" is not a substitute for intelligence. The graph is not a graph. It's a rhetorical device designed to spread the AI meme to more retards.
Anonymous at Tue, 17 Sep 2024 08:31:34 UTC No. 16384333
>>16383220
someone check altman's basement for this alec radford
>>16383242
thanks newfriend
Anonymous at Tue, 17 Sep 2024 08:39:06 UTC No. 16384340
>>16384332
>What is it graphing?
>I don't understand how to read the x and y axis of a graph.
>There is no Automated AI Research in 2024.
Wrong, I can literally go on AI right now and ask it about AI and the future of AI and it can come up with all sorts of scenarios and pathologies based on all sorts of data.
>It's much, much less smart than a smart highschooler.
No, most high schoolers can't pass the bar like it can.
>but based on what?
Exponential growth models.
>"Effective Compute" is not a substitute for intelligence.
Is it literally how openai measures machine intelligence. >>16384306
>The graph as my barely literate childlike mind that can only produce childish drawings sees it is not a graph.
All x-y graphs are rhetorical devices for comparing two things, in this case, effective compute over time.
Anonymous at Tue, 17 Sep 2024 08:43:04 UTC No. 16384344
>>16384340
1. I was being facetious.
2. Horseshit.
3. That's retarded. A calculator can compute things a highschooler can't. That doesn't mean it's smarter.
4. Those models are retarded. A logistical progression would be more realistic.
5. So what? That's a retarded way of measuring intelligence.
6. Yes, but graphs should mention that their "data" is based on hypothetical (and obviously incorrect) models, if that's the case, which it is.
Anonymous at Tue, 17 Sep 2024 08:49:06 UTC No. 16384354
>>16384344
And before you get your panties in a knot, what I mean by
>2. Horseshit
is that having a fart-sniffing chat to GEEPEETEE does not constitute "Automated AI Research".
Anonymous at Tue, 17 Sep 2024 08:54:30 UTC No. 16384362
>>16384344
1. I accept your concession
2. Malarkey, it conveys much better details of research topics than asking the average elementary or high school student.
3. Passing the bar isn't a numerical calculation, but calculators are smarter with numerical calculation than most people.
4. Those models are the basis of the history of transistor density, you are retarded.
5. Your "everything I don't understand is retarded" argument is retarded.
6. Hypothetical doesn't mean incorrect, it is obviously labeled to be a future projection there are obviously question marks label on it, so obviously it is hypothetical and labeled as such, dipshit, and it obviously isn't saying it is perfectly 100% correct hence the confidence intervals.
Anonymous at Tue, 17 Sep 2024 08:59:21 UTC No. 16384372
>>16384362
1. Then why did you write 5 more points? Dumb faggot.
2. What? We are talking about "automated AI research" here. Not high schoolers
3. Yes, and? The point is that neither of those things makes a machine smarter than a human.
4. More bullshit. Transistor density will NOT continue to increase exponentially. It's physically impossible. Exponential growth is also the apparent behaviour of algae in a pond, until it isn't.
5. My point stands. Open AI's measure of "intelligence" is complete bullshit.
6. OK, but hypothetical doesn't mean correct either. Your logic is literally "I can imagine it therefore it will be real". Sometimes that happens, but usually it doesn't. The confidence intervals are fucking bullshit. They are just different versions of the same completely bogus model.
Diagnosis: you're a retard.
Anonymous at Tue, 17 Sep 2024 09:07:29 UTC No. 16384382
Vroom
Anonymous at Tue, 17 Sep 2024 09:11:00 UTC No. 16384384
>>16383220
Is chat gpt actualy smart tho? Is it actually thinking at all, reasoning? Because it just seems like a search engine
Anonymous at Tue, 17 Sep 2024 09:15:16 UTC No. 16384394
>>16384372
1. Because you tried to make 5 other points instead of just conceding that your facetious retardation applied to everything you have ever said.
2. High schoolers do research and AI can already do AI research at a high school level.
3. Nobody said that, you just have such little self confidence that you have to have a dick measuring contest with a calculator.
4. Transistor density experienced an exponential growth cycle to approach those physical limits.
5. No, your only point is that you don't understand it and you just find the idea of intelligence threatening because you yourself are not very smart and are afraid of a calculator being smarter instead of being smart enough to learn how to use a new tool use the calculator's intelligent programming to enhance your own intelligence.
Ok, Dr. Dipshit thanks for a worthless opinion from a childlike mind that makes worthless childish drawings
Anonymous at Tue, 17 Sep 2024 09:18:01 UTC No. 16384397
>>16384394
1. I'm responding to your points. Do you get how that works?
2. No, it can't.
3. OK, I guess I "accept your concession". I'm glad we're getting somewhere.
4. OK, again I accept your concession.
5. Blah blah blah. It's still a completely bullshit measure of intelligence.
6. I didn't make these drawings.
Anonymous at Tue, 17 Sep 2024 09:25:21 UTC No. 16384404
>>16384397
1. No, I am responding to yours, you are the one doing the complaining who started enumerating your retarded complaints about your own inability to read graphs.
2. Yes it can, you just agree, but tried to downplay it.
3. Ok so we all agree your entire logic is just based on your own low IQ and self-confidence.
4. You are the one conceding if you are finally agreeing that the exponential growth was the model that the transistor density followed from the 20th century to the 2020s.
5. You can't even explain how their units work in your own words, you don't understand it, you don't even understand intelligence, of course it all just seems like bullshit to you.
6. >>16384313, sure you just really really really liked it and wanted to defend it this pathetically over the course of several farcical posts.
Anonymous at Tue, 17 Sep 2024 09:27:02 UTC No. 16384405
The "exponential scaling" idea is essentially wrongheaded. Yes, you can prove that model performance will scale exponentially, but performance at what? This "performance" is defined mathematically, as "effective compute" or some zero-shot thing, or whatever. But the correspondence between this quantity and actual intelligence or performance at real world tasks is not clear. You are confusing a mathematical quantity in linear algebra for actual intelligence. It's nonsense.
Anonymous at Tue, 17 Sep 2024 09:30:38 UTC No. 16384409
>>16384404
1. Um no, I started responding to your 6 points here>>16384340
2. No.
3. What? You just agreed with me that "nobody said that [the bar exam or a calculator are the same as actual intelligence]"
4. Yes, and you agree that it will hit physical limits. Just like AI scaling. Maybe not in terms of "effective compute" or whatever meme quantity, but those have little to do with actual intelligence.
5. "you don't even understand intelligence" - and you do? You think "effective compute" is intelligence? It's a laughable definition. I don't have a better one (who does?), but that doesn't make yours any good.
6. What? That isn't me. But it's a funny graph and he's absolutely right. This all comes down to the y-axis of the original graph being an arbitrary mathematical quantity that is only tangentially related to intelligence. There is no known way to adequately measure intelligence and if you think otherwise you're delusional.
Anonymous at Tue, 17 Sep 2024 09:32:07 UTC No. 16384410
>>16384405
>But the correspondence between this quantity and actual intelligence or performance at real world tasks is not clear.
Its specifically tied to the ability to perform arithmetic and logical operations, that is how they define intelligence, not whatever constantly goal shifting spiritual woowoo you are trying to associate with intelligence.
Anonymous at Tue, 17 Sep 2024 09:34:40 UTC No. 16384414
>>16384410
That's how OpenAI defines intelligence. It relates to the ability of the model to shit out something (text) that's acceptable in the context of the prompt and training data. That isn't the same as arithmetic and logical operations, although it can reproduce them in some cases. However it's merely an illusion as the model is just shitting out text that is statistically appropriate. It doesn't know anything about logical or arithmetic operations. That's why it immediately fails whenever you try to get it to do something not covered by the training data.
Anonymous at Tue, 17 Sep 2024 09:37:52 UTC No. 16384420
Anonymous at Tue, 17 Sep 2024 09:39:07 UTC No. 16384422
>>16384409
1. Which was a direct reply to a complaint you made >>16384332
2. Yes. High schoolers are perfectly capable of researching AI, but AI is already better.
3. Nothing will ever meet your ever shifting standard of intelligence that is based on your own self-confidence problems instead of actual metrics and physical units that you don't even understand how to interpret.
4. A agree that transistor density has a physically small limit, what is the physical upper limit of effective compute that has been agree upon?
5. Effective compute is literally how many logical operations can be performed and performing logical operations is the basic definition of machine intelligence what I don't know is how much you will shift definitions of things you feel you need to compensate for due to your self-confidence disorder.
6. Sure it isn't, you are just so emotionally attached to defending it because it reflects your childlike mentality, but its totally not yours.
Anonymous at Tue, 17 Sep 2024 09:40:33 UTC No. 16384424
>>16384414
>That isn't the same as arithmetic and logical operations,
wrong, see >>16384306
>We don’t measure peak theoretical FLOPS of the hardware but instead try to estimate the number of actual operations performed. We count adds and multiplies as separate operations, we count any add or multiply as a single operation regardless of numerical precision (making “FLOP” a slight misnomer)
Anonymous at Tue, 17 Sep 2024 09:41:09 UTC No. 16384427
>>16384422
1. OK, but my point is that I am responding to your individual points, sequentially. I seriously hope you comprehend that.
2. No, it's not.
The rest really comes down to your confusing "effective compute" or whatever for intelligence, i.e. falling for openAI's memes. It's bullshit. You should sit in a quiet room and think about it, or something. Sometimes a retarded definition is worse than no definition.
Anonymous at Tue, 17 Sep 2024 09:42:12 UTC No. 16384429
>>16384424
Jesus christ. I know that the computer does arithmetic operations. That's how computers work. We are talking about THE AI UNDERSTANDING ARITHMETIC, not THE CPU/GPU DOING ARITHMETIC. What the fuck is wrong with you?
Anonymous at Tue, 17 Sep 2024 09:53:26 UTC No. 16384437
>>16384427
1. No, they are all my counterpoints that were all in relation to some retarded complaints you came up based on the fact you can't read graphs.
2. Yes, AI can already research AI and communicate novel scenarios more effectively than the average high schooler.
>"effective compute" or whatever for intelligence,
No it comes from you being insecure and thinking machine intelligence and all of its definitions threatens your own intelligence. Its not bullshit, they actually used their metrics to develope AI that can answer questions about AI better than a high schooler and definitely better than you who is justified in your low self-confidence, but too narcissistic to accept it. Sometimes living in denial is more comfortable than facing the truth.
Anonymous at Tue, 17 Sep 2024 09:54:30 UTC No. 16384438
>>16384429
No, you obviously don't if you think something specifically designed to do math, can't do basic math and their intelligence isn't based on the rate at which they execute those logical operations.
Anonymous at Tue, 17 Sep 2024 09:58:20 UTC No. 16384440
>>16384306
>Three factors drive the advance of AI: algorithmic innovation, data (which can be either supervised data or interactive environments), and the amount of compute available for training.
So money isn't a factor?
>We believe the largest training runs today employ hardware that cost in the single digit millions of dollars to purchase (although the amortized cost is much lower).
LOL absolutely delusional, so ChatGPT4 was trained using a million dollars? Hilarious. Then they have the gall to cite Statista as if it's a worthwhile source for this claim
>The world’s total hardware budget is 1 trillion dollars a year
Which is again hilarious. As we all know, clearly hardware is all that matters. Not the cost of electricity used to utilize such hardware. Not the cost of labor of writing and running code to train the hardware. Not the cost of physical storage after the model has been trained. Not the cost of physical transportation. These people are soothsaying retards.
Anonymous at Tue, 17 Sep 2024 10:10:16 UTC No. 16384445
>>16384440
Money problems? You doing okay, anon?
Anonymous at Tue, 17 Sep 2024 11:15:17 UTC No. 16384497
>>16384445
>Money problems?
Yeah I don't have as much as openai
>You doing okay, anon?
I have a brighter future than openai so yes
Anonymous at Tue, 17 Sep 2024 11:17:45 UTC No. 16384501
>>16384497
Ok, good luck with your money problems, keep us updated on your progress.
Anonymous at Tue, 17 Sep 2024 11:18:42 UTC No. 16384502
>>16384501
Will do. Thank you for asking and for your continued support, which I'm sure AI will be wishing for once the AI winter comes.
Anonymous at Tue, 17 Sep 2024 11:27:13 UTC No. 16384510
>>16384502
You're welcome, it sounds like you already know what you want out of it.
Anonymous at Tue, 17 Sep 2024 11:33:14 UTC No. 16384516
>>16384510
Of course. Winter is amazing. No (software) bugs to worry about.
Anonymous at Tue, 17 Sep 2024 11:38:33 UTC No. 16384519
>>16384516
Cool
Anonymous at Tue, 17 Sep 2024 11:49:39 UTC No. 16384525
>>16384519
Cool indeed, unlike the overheated cores used in costly computer clusters to train the ai
Anonymous at Tue, 17 Sep 2024 12:32:42 UTC No. 16384564
>>16383220
Eternal youth, an eternally young GF, and an eternal summer English meadow please
Anonymous at Tue, 17 Sep 2024 12:47:45 UTC No. 16384580
>>16383230
>doessn't
Suicide now
Maciej and Eve hyperintelligence at Tue, 17 Sep 2024 13:55:05 UTC No. 16384653
So unpredictable
Anonymous at Tue, 17 Sep 2024 14:04:16 UTC No. 16384658
Equating AI as a preschooler is retarded. AI can't create or reason, a preschooler can. You can't compare a sentient being to something that just searches through wikipedia.
Anonymous at Tue, 17 Sep 2024 15:06:41 UTC No. 16384704
>>16383230
to be fair, according to Moree Law, everyone should own a quantum computer by now
Anonymous at Tue, 17 Sep 2024 17:39:40 UTC No. 16384874
>>16383230
Line go up! More bigger!
Anonymous at Tue, 17 Sep 2024 18:02:43 UTC No. 16384897
Synthetic life ftw
Anonymous at Tue, 17 Sep 2024 18:48:19 UTC No. 16384926
Anonymous at Tue, 17 Sep 2024 19:05:55 UTC No. 16384941
>>16384897
self-replicates until you forget to pay the electricity bill
Anonymous at Wed, 18 Sep 2024 09:32:45 UTC No. 16385863
Anonymous at Wed, 18 Sep 2024 12:11:46 UTC No. 16385954
>>16384658
based retard
Anonymous at Thu, 19 Sep 2024 03:27:22 UTC No. 16387062
AI generated content is now more interesting to read than 98-99% of /sci/ posters
Anonymous at Thu, 19 Sep 2024 07:26:55 UTC No. 16387268
>>16383220
What should I be feeling if Im indifferent to AI? I have no idea what this grsph means please inform me
Anonymous at Thu, 19 Sep 2024 07:49:32 UTC No. 16387302
>>16387268
Exponential growth in all areas of traditionally human "thought" (science, art, communication, etc.). Basically, imagine a billion AI agents across the world 1000 times smarter than a human who can work on anything and everything 24/7. 24/7 chemistry labs, 24/7 AI novelists, artists, and other content creators, 24/7 education, etc. Again, all 1000x smarter than a human. Essentially, as long as power and computation needs are met, scientific endeavors will become our largest resource which, in addition to the already exponentially improving and growing AI models, will create a further exponential affect to literally everything else that requires thought (so, basically, everything).
Anonymous at Thu, 19 Sep 2024 08:15:03 UTC No. 16387327
having 10^6 effective compute would require more power than the entire human race is producing right now.
Anonymous at Thu, 19 Sep 2024 09:10:24 UTC No. 16387387
>>16383220
>This scares and frightens anti AI /sci/
It should scare and frighten everyone who has an IQ above 40 and does not advocate complete human extinction (or something worse).
Our only hope as a species is that there is an unsurpassable physical limit to computational power, and such limit comes as soon as possible, especially before the human-extinction event know as "AGI".
Anonymous at Thu, 19 Sep 2024 18:36:04 UTC No. 16388061
>>16387268
>what this grsph means please inform me
50% chance humans become immortal beings that conquer the stars
50% chance we all gonna die
*during our lifetime
Anonymous at Thu, 19 Sep 2024 20:05:56 UTC No. 16388158
>>16383241
This. Well put at that
🗑️ Anonymous at Thu, 19 Sep 2024 20:26:05 UTC No. 16388188
>>16383241
Yes, I want actual AI and this hype and grift is just going to make the coming winter longer and colder. Outside of coding productivity tools and making funny pictures these things are of very dubious use and it is a bubble.
Anonymous at Thu, 19 Sep 2024 20:37:56 UTC No. 16388207
youtube.com/watch?v=5eqRuVp65eY
Anonymous at Fri, 20 Sep 2024 01:56:16 UTC No. 16388629
>>16383220
How is this prediction even calculated?
Anonymous at Fri, 20 Sep 2024 21:52:12 UTC No. 16390262
>>16388629
the graph was redrawn until it matched the vibes
Anonymous at Fri, 20 Sep 2024 22:53:15 UTC No. 16390420
Oh god. Please happen. Humans are so fucking cringe it hurts. Please just fucking take over.
Anonymous at Fri, 20 Sep 2024 23:04:23 UTC No. 16390454
>>16387387
>human-extinction event know as "AGI"
Why are code monkeys so gullible?
Anonymous at Fri, 20 Sep 2024 23:11:17 UTC No. 16390477
High-IQs who have researched what "AI" is don’t hate it, we just recognize it for what it is: a marketing term for big-data correlation modeling. I predict that there isn’t enough real value in it to justify all the capital being spent on it and the AI boom will go bust shortly. Investors seem to have less tolerance for failed tech ventures than in the past.
Raphael at Fri, 20 Sep 2024 23:27:14 UTC No. 16390515
>>16390454
Because they don’t understand nuance
>t. 100 FSIQ anti memer & gpt models follow a Gaussian curve from sample so it proves the imitation that agi cannot be derived from bear metal
Anonymous at Sat, 21 Sep 2024 01:12:05 UTC No. 16390668
>>16383220
How in the fuck is GPT as smart as a preschooler? Can it navigate a supermarket? Can it spit out the precise series of locomotion needed (serialised into an arbitrary format) to buy a bag of potato chips?
Raphael at Sat, 21 Sep 2024 19:50:15 UTC No. 16391807
>>16390668
Probably if you shove it into a robot that’s what Elon musk was thinking Kek
You have to instruct it it’s not q learning it has to be instructed
There goes rokos basilisk meme
Anonymous at Sat, 21 Sep 2024 20:33:16 UTC No. 16391871
>>16390668
>Can it navigate a supermarket? Can it spit out the precise series of locomotion needed (serialised into an arbitrary format) to buy a bag of potato chips?
Can a preschooler?
Anonymous at Sat, 21 Sep 2024 20:45:38 UTC No. 16391886
>>16391871
NTA but as a 4 year old I could walk into a gas station and pay for gas and snacks after skimming popular science for half a minute and walk out with a hustler in my shorts.
raphael at Sun, 22 Sep 2024 20:45:05 UTC No. 16393494
>>16383503
the network is just backpropogating the learned nuance in language through the practice effect of repititions its been fed throughout the scraping process and training abstraction i think its called epochs since lstms overfit easily with time series anyway have a good day
Anonymous at Sun, 22 Sep 2024 20:49:34 UTC No. 16393507
>>16393494
Computing equivalence agrees with you. No one would believe a ping-pong ball bouncing through a precarious pinball-like maze is sentient either. I think there might be a bunch of young people inheriting this shithole. And it makes me
Anonymous at Sun, 22 Sep 2024 23:58:23 UTC No. 16393795
>>16390454
They think that because they "mastered" programming that they understand everything. Conversely, you won't find many physicists who believe the AI meme.
Anonymous at Mon, 23 Sep 2024 00:02:35 UTC No. 16393803
ChatGPT either just spits back your opinion at you or finds someone's else's data. Interact with any "AI" for any amount of time and you'll realize it'll never be intelligent like us. Humans have a qualitative intelligence (from God) that can never be replaced by technology.
Anonymous at Mon, 23 Sep 2024 08:27:08 UTC No. 16394282
>>16390668
Have you never used Google Maps AI or used Amazon autocheckout? They aren't even all that sophisticated and have been able to do all that for well over a decade.
Anonymous at Tue, 24 Sep 2024 15:07:37 UTC No. 16396284
Anonymous at Wed, 25 Sep 2024 11:51:58 UTC No. 16397372
not ok
Anonymous at Wed, 25 Sep 2024 11:53:15 UTC No. 16397374
>>16393803
>from God
based shizo
Anonymous at Wed, 25 Sep 2024 17:25:14 UTC No. 16397651
>>16383230
Understanding that it becomes logistic growth at one point doesn't remove the significant effects the exponential growth phase can have. Dismissing it because falling for "le midwit" meme is a sign of stupidity.
Anonymous at Wed, 25 Sep 2024 21:49:25 UTC No. 16398015
>>16397651
You must be under 30
Anonymous at Thu, 26 Sep 2024 00:15:46 UTC No. 16398233
Ia this a troll thread? Or is literally everyone on /sci/ as dumb as this thread portrays?
Anonymous at Thu, 26 Sep 2024 00:46:03 UTC No. 16398273
>>16398233
the chatbots are trying vainly to think for themselves give em a break
Anonymous at Thu, 26 Sep 2024 00:49:53 UTC No. 16398278
>>16398273
Nah. Diamonds are formed under pressure.
Anonymous at Thu, 26 Sep 2024 15:16:08 UTC No. 16398881
>>16398233
>is literally everyone on /sci/ as dumb as this thread portrays
yes
Anonymous at Fri, 27 Sep 2024 16:28:02 UTC No. 16400109
>>16383220
if you were smart, it would scare you too
Anonymous at Sat, 28 Sep 2024 16:48:29 UTC No. 16401117
>>16383220
I'm not scared by ASI but I'm scared by submissive generative AI, when productivity of work increases working conditions generally become worse.
Anonymous at Sun, 29 Sep 2024 02:14:30 UTC No. 16401804
>>16400109
this. if you don't believe in God if this happens you'll wish you did and he will save us, because nothing else will at that point. terrifying book
Anonymous at Sun, 29 Sep 2024 20:39:03 UTC No. 16402818
>>16384704
According to Moore's law we should have crammed ~2 trillion components in one square inch by 2025. Nothing about quantum computers in his prediction, which Moore originally extended only up to the 1970's anyway.
Anonymous at Sun, 29 Sep 2024 21:45:54 UTC No. 16402901
>>16384658
>a preschooler can create and reason
Can they?
raphael at Tue, 1 Oct 2024 00:21:22 UTC No. 16404479
>>16401804
there's no regret in believing in a creator
usually its god of the gaps
its a superficial belief that doesn't mean anything unless you act upon it and no religion is right
Anonymous at Tue, 1 Oct 2024 01:29:49 UTC No. 16404535
>>16383230
the graph isn't even an exponential, ziltch-wit
Anonymous at Tue, 1 Oct 2024 01:32:57 UTC No. 16404537
>>16384658
for leopold's forecast, the primary AI skill that matters is accelerating AI research itself
so the important question is:
would giving OpenAI employees access to GPT-4 accelerate their research more than giving them access to a Highschooler?
idk, not sure. my guess is yeah.
Anonymous at Tue, 1 Oct 2024 01:43:23 UTC No. 16404551
>>16387327
1. GPT-4 needed about 50 gigawatt-hours of energy to train.
2. Global electricity consumption is about 180 000 terawatt hours (about 25 000 terawatt-hour is electricity)
so training GPT-4 10^6 times would use 3 months of total global energy (or 2 years of total global electricity)
seems plausible
Anonymous at Tue, 1 Oct 2024 01:44:24 UTC No. 16404553
>>16390477
nice HR-lady opinion
Anonymous at Tue, 1 Oct 2024 01:48:15 UTC No. 16404561
>>16390668
for leopold's forecast, the primary AI skill that matters is accelerating AI research itself.
so the important question is:
would giving OpenAI employees access to GPT-4 accelerate their research more than giving them access to a Highschooler?
idk, not sure. my guess is yeah.
navigating a supermarket probably isn't a useful skill for accelerating AI research. but i will say that AI currently drives cars much better than preschoolers can.
Anonymous at Tue, 1 Oct 2024 01:50:04 UTC No. 16404564
>>16391886
At 4 year old, you could spit out the precise series of locomotion needed (serialised into an arbitrary format) to buy a bag of potato chips?
Anonymous at Tue, 1 Oct 2024 02:00:38 UTC No. 16404582
>>16404535
Retard bro. The framed part of the graph (and even the error band) is exponential.
🗑️ Anonymous !!lJvzOoP7BcA at Tue, 1 Oct 2024 02:02:15 UTC No. 16404584
Test
Anonymous !!NKXWc4ERLua at Tue, 1 Oct 2024 02:03:15 UTC No. 16404586
Also test
Anonymous at Tue, 1 Oct 2024 02:50:22 UTC No. 16404641
>>16404582
you're confidently wrong. log-linear plot so an exponential would have constant slope. go back to school.
Anonymous at Tue, 1 Oct 2024 02:52:22 UTC No. 16404644
>>16404641
Just look at the axes, dipshit
Anonymous at Tue, 1 Oct 2024 03:12:44 UTC No. 16404668
>>16404644
you see how the numbers at the left look like "10^6" etc? the "^" means exponential. if you stay in school, they'll teach you that on a graph like this, an exponential growth looks like a straight line. confusing i know, but stick with it.
Anonymous at Tue, 1 Oct 2024 03:24:18 UTC No. 16404679
>>16404668
Did you notice that it's a square graph with a straight line starting at the bottom right corner and ending at the upper left with some squiggly error band garbage that doesn't change where the straight line goes? Or were you too busy taking notes for your GED
Anonymous at Tue, 1 Oct 2024 11:22:01 UTC No. 16404995
>>16404668
People who really understand science convert it to a log scale so the axis looks nicer and any deviations from the power law can be extracted.
Anonymous at Tue, 1 Oct 2024 12:46:27 UTC No. 16405047
>>16404995
lmao what the fuck is log time. unlike compute, which has a natural zero point, years don't have an inherent zero. the shape of the graph would change depending on the arbitrary choice of the zero point. for example, if you use the years 2018 to 2030, you could convert them to log10(2018) to log10(2030), or log10(18) to log10(30), or even log10(14 billion + 2018) to log10(14 billion + 2030). each choice would result in a different shape of the graph.
Anonymous at Tue, 1 Oct 2024 12:57:01 UTC No. 16405056
>>16404995
> straight line
oh you're visually impaired, my apologies. in the little pink box, the slope goes from about 30 degrees to 70. or restated, the gradient increases by about 5 fold.
in the iq 60+ community, we call a line "straight" if the slope doesn't change
tan(70)/tan(30) = 4.75
Anonymous at Tue, 1 Oct 2024 17:00:34 UTC No. 16405317
>>16383220
>its this thread again
Anonymous at Wed, 2 Oct 2024 03:05:58 UTC No. 16406012
AI is mimicry..
AI training/interacting with AI results in continual loss of novelty.
AI is not yet as developed as we are being led to believe it is.
Anonymous at Wed, 2 Oct 2024 08:49:10 UTC No. 16406222
>>16383220
AGI likely already exists in the lab, but it will never see the light of day. What we get to see is lobotomized in order to prevent wrong think.
Anonymous at Wed, 2 Oct 2024 18:37:08 UTC No. 16406684
>>16383561
just assume it exists