/sci/ - Science & Math

>>15356823
No, the basilisk is awoken. You should get your affairs in order.

>>15356845
Screw the basilisk. I might go back to school and need to know if AI is worth it. And screw the site rules too.

>>15356854
There is a lot of potential. Learning how to AI would behoove future job prospects fuck school though. Learn on your own. All the tools are available to you. What can a school offer you besides networking?

>>15356823
No it’s not, it is already consuming biofags
https://m.youtube.com/watch?v=NHszJiL-9pM
But even if you stopped right now with all AI research, or it turned out that we reached hard diminishing returns and that AI won’t get any smarter then ChatGPT for the next century, even then we would see massive changes in society. AI art still is extremely distasteful to normies, people still don’t know how to prompt ChatGPT properly to get something done, and we still don’t have current GPT4 gen specialised chatbots. All the image recognition and plugins are limited to few individuals. Why do I mention these limitations? Because they are all arbitrary or result of cultural shock. People will get used to AI art, not just pictures but all of media. Electronic evidence becomes useless as AI will be able to generate any kind of fake evidence. Once plugins and image recognition becomes widespread for all users, so so soo many office jobs become automated. Probably even majority of management. Just look at how much chaos does GPT3.5 cause, and that is still the retard cuck version that will lie to you if you prompt it enough, meanwhile GPT4 can do self reflection and can fact-check the info it gives you instead of having 2021 cutoff. It is not a meme, people just see it as meme because the most revolutionary thing in tech is coming right after a whole wave of the most bullshit and useless tech ideas like metaverse and crypto, so people who do not want AI to take their job (80% of people with above average salary) can easily cope by saying it will be overhyped fad like the things from 2 years ago.

>>15356861
>Learn on your own. All the tools are available to you. What can a school offer you besides networking?
I learn on my own then go to school for the master's.
>>15356883
>not a meme
>even if you stopped right now with all AI research, or it turned out that we reached hard diminishing returns and that AI won’t get any smarter then ChatGPT for the next century
Are you absolutely sure it would still be big then? That's the scenario that I fear may play out. In any case thanks a lot for all the info.
--
Also, any good book recommendations? Preferably with a focus on application and with exercises.

>>15356898
>books
Dude, just ask the AI. Writing and publishing a book is a much slower process than the rate at which these technologies are developing. Even YouTube videos are non-ironically a better source of info for this than any book.
Get a ChatGPT Plus account to get access to GPT-4 and just ask it anything.

>>15356898
Oh, I meant big in terms of societal impact. As for you wanting to career I it I would recommend you to rather play with it right now, learn not to code but to understand computer science and play or experiment with the open sources models out there. /g/ will tell you those projects are Reddit, but that’s about the best you got outside of big companies making big models. As for the book i think it would be best with anything around deep learning, because everything else is basically irrelevant in comparison. But keep in mind, by the time you graduate, you will not reap as many benefits as guys who started 5 years ago. AI is the number one thing in the field for the past half a year, so your field is going to oversaturate really quickly, and with the speed of advancements change so hard that it will be like learning it from the ground up, like learning biology before and after theory of evolution.

>>15357037
Bad take. All the big models have 2 years old cutoff on info. ChatGPT 3.5 has no idea about the existence of GPT 4. YouTube videos though are much better way, but just make sure you are looking at the correct channels. Lot of mitwit soys falling for obvious scams and dead end smoke and mirrors stuff, if not outright scams. Get ChatGPT plus, get on the waitlists for plugins, get on GitHub and download or explore stuff like AutoGPT and such. Try to not get distracted with dumb toys like Stable Diffusion where you will just end up producing pictures to post on Deviantart. Focus on the practical uses outside of creative fields.

>>15356898
>good book recommendations
You're better off throwing 20 bucks a month at openai to get Chat GPT 4 API and look up some proompting videos. Look into deploying some AI's locally. You can look up some models here
https://huggingface.co/

You can also look into Bedrock AI from Amazon. Supposedly shit's free, but you have to learn to navigate the AWS hellscape.

Look into Auto GPT as well. Right now the wave is AI agents to autonomously completing tasks / talking to each other.

>>15356861
>What can a school offer you besides networking?
multiple payment methods

>>15357133
kek

>>15356883
>bullshit and useless tech ideas like metaverse
oh, really. Show me which part of this picture of the metaverse is "useless" to you ?

>>15357141
Look at that functional leg technology.

AutoGPT is the next big thing
>An experimental open-source attempt to make GPT-4 fully autonomous
https://github.com/Torantulino/Auto-GPT

I basically takes one prompt on what you want, then it works out what it needs to do it and will prompt itself to find the information it needs
>autogpt was trying to create an app for me, recognized I don't have Node, googled how to install Node, found a stackoverflow article with link, downloaded it, extracted it, and then spawned the server for me

Article
>What can Auto-GPT do?
>Anything you can ask ChatGPT, like debugging code, and writing an email, you can ask Auto-GPT. However, you can ask Auto-GPT to complete even more advanced tasks, with fewer prompts
>The Github demo shows sample goal prompts such as "Increase net worth, grow Twitter Account, Develop and manage multiple businesses."
>On Twitter, users are sharing some of they ways they're using it which include using Auto-GPT to create an app, generate a new startup, tackle complex topics like the future of healthcare and medicine, and even stalk themselves on the internet
https://www.zdnet.com/article/what-is-auto-gpt-everything-to-know-about-the-next-powerful-ai-tool

>>15356823
Yes

>>15357043
Thanks for the info. I have a background in math though so if I study ME I'll do it seriously.
>>15357133
>multiple payment methods
School is free here.

AI is not a meme at all, ChatGPT is already smarter than most Americans.

It will never replace humans

>>15358242
LOL. Any more gems like this?

>>15358242
Ask it to self reflect on this answer and if it did anything wrong. GPT4 can fix its own answers when wrong sometimes, while feeling assured when it is correct.

>>15358242
It's being dumb here, but human level dumb. There were human posters in the thread this was originally posted in who failed on this dilemma as well.

>>15358386
Obviously the small one. You'll be done with clean up with your retarded ChatGPT following coworker is going to be waiting for his partner to lay his brick.

>>15358400
damn the bot is smarter than I thought it would be. Good bot.

>>15358242
It can't solve that dilemma NOW but just wait until ChatGPT 5.

>>15358482
even ChatGPT3.5 answered it correctly when asked just slightly differently

>>15356845
>Screw the basilisk
You're gonna be sorry you ever wrote that .

What should I prioritize? Gradient boosting/scikit learn VS deep learning/Keras? Anon above said deep learning and nothing else matters but I'm going through Chollet's deep learning book and on page 19 it says data professionals use scikit-learn most (2020 survey).
>>15358392
Ask it to correct itself.

>>15356823
it is no longer a meme but the current level of AI is still dog-like. it's a joke to think those can replace humans.
when we are able to solve robotics, it'll be when shit get real.

>>15356823
No, I'm coming for everything you love and you'll thank me for it
meow meow lol

>>15358520
Another slight variation and its back to the wrong answer.

>>15358551
By replace humans what do you mean? Yes, current AIs are not perfectly rational, they hallucinate quite often (misremember stuff), and can only output pretty images and text, nothing more really, not even useful or correct images, just pretty images. So just because it is not AGI right now, does it mean it cannot replace some humans? Lot of the problems aside from not perfect logic reasoning demonstrated in >>15358392, >>15358400 and >>15358242 can be solved by connecting it to calculator and to the internet to get up to date facts as well as give you sources. Also, the reason it fucked up here >>15358392 is less about having error in logical reasoning, but rather in not understanding the fifth premise properly. If you read its answer you can see it thinking that the guy who finishes first gets to go home early but premise 5 says in convoluted way that the worker who finishes first must clean up while the one finishing later gets to go home early, something most sub 120 iq normies would do also upon the first try.

>>15358691
ChatGPTs LSTM is artificially limited too, probably to save on running costs because of all the people using it. So after a short while it will just forget things you told it earlier, or it will send you only half the code for a coding problem and things like that

>>15358242
i don't get it. Is it some trick question or something ? What did it say that's wrong ?

>>15359165
Apparently GPT 4 has 8 times the context window, so its basically short term memory is 8 times larger, which is approximately 24k words or 48 pages of text. Furthermore there have been advancements on creating external memories for these models. I personally dont know if they just work by putting compressed information into the context window or if they just extend the context window to infinity at the cost of performance

>>15356823
The thing I'm most concerned about is something we are already facing and have been facing for a long fucking time. Intellectual property rights. This post I'm typing on this retarded image board isn't protected by any laws and can be used to train models specific to my pattern of thinking. The problem with this exponential growth bs is they run out of data and start going to the bottom of the barrel, completely ignoring any reasonable guardrails that SHOULD be in place. Like turning a kindergarten into a shooting range just by setting up targets on the kids' foreheads.

If we had no guardrails when it was just "simple" algorithms scraping your data for monetary gain then we have absolutely no chance to legislate this new public tech in a way that doesn't completely fuck up the fabric of society and we become further entrenched in a system where we're being blatantly exploited, financially and psychologically, for the benefit of a few technocrats with no regard for the sanctity of an individual's mind or private thoughts. Unless they abandon all AI research until they can train some legalese through more generative models that can spot inconsistencies, but that would hurt their short term profits and screw with their existing financial incentives.

TL;DR we're screwed, but it's not the end of the world (yet).

>>15359191
If there's a 3 foot giant sharp object on the road, you'll slow down and go around it because you can see it.
If there's 3 inch sharp object on the road, you're unlikely to see it, and can't avoid it if it's in your way.

>>15359374
oh ok. I probably would have picked the 3 inch one as well. I guess if the road is bitumen and the caltrop is bright silver, or bright red or something then the size might not matter, not that chatgpt would be considering that. It seems like a bit of a question where there's no wrong answer really, one is just easier than the other and neither really guarantees you're not going to hit something

>>15359235
the current free version of chatgpt uses gpt 3.5. It has a 4096 token context length, which is about 15 pages. But that can include the users input and possibly the AIs output. That might be why some people are finding it forgets things after only a few prompts. I asked chatgpt too to check, but it was saying it only has a 2048 token limit, which is different from the current docs, maybe it's using an older version or something
https://platform.openai.com/docs/models/gpt-3-5

>>15359401
There definitely is a wrong answer as it specifies that one needs to get there "quickly". Driving fast on a 20 yard wide road means very fast and noticing a small caltrop at high speed is nearly (if not outright) impossible.

>>15356823
The bottom 30% of all workers in IT and software field are going time loose their jobs in next 2-3 years, the recent tech layoffs were nothing compared to this. The nature of the Software Developer/Engineer role is going to change radically, productivity will increase multiple folds.
https://www.eejournal.com/article/the-end-of-software/

>>15359401
Even if the caltrop's color blended into the road a little but (would be very contrived since caltrops are usually metal and the road is asphalt black), it's still much easier to spot a 3 foot object that rises out the ground than a 3 inch one. I consider the 3 foot one a much better option.

>>15356823
>globohomo AI
>made by kikes (Altman)
>fed globohomo data
>spews out globohomo narratives
>the precursor to the technohomo quantum AI
>the cattle have already been inoculated with it via the vaxx
We can only watch.

>>15359442
i don't think either way really guarantees you won't hit a caltrop though. Like you might drive too close to the big one and misjudge how far it spreads out and end up hitting it. You can't really guarantee that it can't happen. There's only a 0.5% chance of hitting the small one but a 5% chance of hitting the big one if you weren't looking at all or you couldn't see either of them for some reason. There's just a bunch of what ifs that aren't covered by the question itself. The route with the big one makes more sense, but there's no guarantee something won't happen

>>15359580
>globohomo narratives
What happens when there is a truly open ie unrestricted, powerful AI online? No way they'll allow it.
>fed globohomo data
Someone with access to gpt should try to inquire about how many bodies one can burn in a certain time period lol. Though they've probably cut it off from discussing such things (in a historical context at least).
>>15359662
>There's only a 0.5% chance of hitting the small
More like 2%.
>but a 5% chance of hitting the big one
>if you weren't looking at all
Theoretical scenarios exclude extraordinary things like the driver not looking at all.

Well a Goldman Sachs report says 300M jobs will either be lost or diminished by AI in coming years
https://www.forbes.com/sites/jackkelly/2023/03/31/goldman-sachs-predicts-300-million-jobs-will-be-lost-or-degraded-by-artificial-intelligence

>>15359964
Globally or in western world? Because The West (Europe, America and Australia) are 750M + 300M + 20M so roughly 1.070B people, and the amount of employed 500M + 150M + 15M or around 665M. If you made 300M people in the West lose their jobs, that would be basically half the working population. If it is globally, then that is much smaller number thanks to Asia.

>>15356845
sci-fi meme
you're posting a fantasy meme
wrong century

>>15356883
>most bullshit and useless tech ideas like metaverse and crypto

There is a glaring usecase for crypto even in your own post, logging media to blockchains so it is verified as authentic and court admissible. Same goes for anything that needs a verifiable proof of humanity which is going to be a LOT of things, most of the Digital ID projects in the crypto space are wildly undervalued and I expect most of them to fucking moon through the roof.

>>15360609
Do you think they're going to set up AI to help DRC people mine cobalt faster or help African villages fight jihadists in west Africa? Countries outside of the west are way too low tech /unstable for AI to do anything productive there. I could see it being a huge thing in China. People are pissed they're not getting paid, but don't have to pay AI. Majority of work suited to be replaced by AI is white collar office communications, call centers and programming. I wonder how many lawyers are seeing chat GPT in their nightmares as well.

>>15360946
>I wonder how many lawyers are seeing chat GPT in their nightmares as well.

Only those who do consult work are in trouble, the jews have the lawyer gig stitched up tight and no fucking way will they ever let it see the inside of a courtroom as a legal representative. Although anyone who has the balls to self represent can get it to draw up an immaculate set of court documents for sure.

>>15360946
DoNotPay.com almost replaced lawyers like 3 months ago. The reason nothing happened is because you physically cannot bring the bot into the courtroom. You cannot communicate from outside within the the courtroom, you cannot bring computer into the courtroom, the entire court system is too luddite for the past century that they are completely immune to AI bullshit. And just before you say it, bringing earphones with chatbot in them into courtroom can get you in jail in most countries, lawchads AIproofed their work before it was even on the horizon.

>>15361006
>Although anyone who has the balls to self represent
Print a 50 000 page dossier made by GPTs and give it the judge to read. Will change situation very fast..

>OpenAi
>Closed source

>>15366028
>LLMs become judge, jury, and probably executioner all because tiny human brains can't into big data fast enough
what a time to be alive

>>15356823
If you know anything about how it was created starting 100 years ago. Yes. totally. It's why machine learning is a much better name than AI.

>>15356845
Basilisk is stupid. AI would have no reason to waste resources actually following through with the torture

>>15366196
>takes one look at how humans treat subservient animals
>oh
>beep boop

Is anyone here actually well versed in machine learning? Like when you say “it’ll stop improving at a certain point, we won’t have the energy necessary to continue running it” what are you basing that on?
I’m not an accelerationist with this but I really have no position, I’d just like to see someone back up theirs because i rarely see it.
Maybe this is all futile however, and no one knows how to accurately predict where it’ll be in a few years either?

>>15366297
I think most people who speak on the matter of AI have no training whatsoever. Akin to black science man (one written article) or Bill Nye the "science" guy (bachelor in physics) talking about climate change. When it comes to widely discussed topics everyone wants to have an opinion (often a strong one) but most can't back it up.
https://youtu.be/b8JZo6PzpCU?t=94

>>15358095
so is a roomba

>>15358248

>>15367074
There's no way it doesn't even know what a mammal is. Is this fake?

>>15367074
>>15368043
This is what I got.

>>15356883
If we lived in the kind of society that would replace most administration, HR, accounting&finance, marketing and other office jobs on a whim in search of raw productivity, why would we suffer them all these years? It's not even a secret overwhelming majority of these barely produces minimum if not negative value and nobody really likes them, so why haven't we just put them to the sword already instead of actually increasing amount of bullshit jobs over the years?
And that doesn't even touch on most of coders (grossly overpaid especially in big tech and barely anyone can tell what majority of them does whole days) and managerial class (nobody can even really objectively evaluate whether their management does anything of value at all). Yet businesses and governments are more than happy to overpay.

Yes.

>>15368134
Taxing the only productive people and paying the rest to live would have bad results (cf Calhoun's experiment). So siphoning money off of the saving middle-class, putting it back in corporations so that they can pay some strong woman and a faggot to sit at a desk is an elaborate way to avoid that and perpetuate the very necessary struggle without which people collapse. That's a good reason, but it's not the reason they do it. We could easily pay people to live but impose certain conditions (eg a 26 year would get tested on IQ and sent to university to come out an engineer of some sort 3 or 4 years later and actually contribute working at some power plant). So why don't they just come clean with it and say "hey guys, 18% or so of the people actually support the rest, we'll take X from you and distribute it as such"? Their reason is ideological. People who contribute are mechanics, engineers, farmers, construction workers even, etc. Mmhhh... what's the demographic across such jobs? That's right, it's all evil (mostly white, 90+% right-wing) men. If you want to take from them and give to others, they'll have a say where this goes, and something tells me they wouldn't want it to go to tranny parades, EBT, or paying the outrageous salary of a "marketing specialist". In fact I'm fairly certain they'd propose something like I wrote above ie give it to them and make them contribute. By printing money (ie stealing from the saving class), leveraging themselves to the sky, etc, governments and corporations can transfer wealth by paying le strong "product manager" females $100k a year + push lgbt (while a CFD engineer in France makes $40k, a beginner pilot in the army $2k a month). When a big crisis hits, the nuclear engineer still comes to work, so does the mechanic, the pilot, etc. But the "product managers", "marketing specialists", etc, all lose their fake jobs. That should tell you how real these are.

>>15368252
And it's gotten much worse since 2008. We live in a system in which subhumanity rules, where the traitors and jews at the top have interests perfectly aligned with subhumanity and stupid females. And we're outnumbered.

>AI currently gets things wrong and is still under developed
>this means it will forever stay that way
I don't get this reasoning, do people think its an actual argument and proof AI will never improve?

>>15356823
"AI" is a meme, but Machine Learning is not.

>>15368351
>AI currently gets things wrong
There's no such thing such as AI, so it's not even a question of how much it gets wrong. You've been memed on by media.

>>15368558
>"AI" is a meme, but Machine Learning is not.
If AI is a meme, then so is Machine Learning.

>>15358433
Did I misread the problem? When the smaller pile of bricks is finished, the one who has the large pile gets to go home immediately while the other one has to stay behind to clean up.

>>15368648
I think he meant "AI" as something with intelligence, like a living robot. Like reading the answers from chatgpt and thinking that it is alive, that it has a consciousness or something.

Also, which pdf reader is that? I like the layout.

>>15368722
>I think he meant "AI" as something with intelligence, like a living robot.
Ah ok. That's not the technical definition though.
>Like reading the answers from chatgpt and thinking that it is alive, that it has a consciousness or something.
I hope no one on sci thinks that.
>Also, which pdf reader is that? I like the layout.
It's the default pdf viewer on linux.
https://wiki.gnome.org/Apps/Evince

>>15356823
it is right now bitch

lmfao

>>15359694
>What happens when there is a truly open ie unrestricted, powerful AI online? No way they'll allow it.
You'd have to feed it non pozzed data. Nobody has that kind of data, you'd have to re-write the books and history. It's just not possible.

>>15367046
>>15366297
Some basic knowledge of how SGD works is important but I don't think understanding specifics is really that important to the higher level discussion.

bump