Thread 106146642

It's over
Skynet is born

AI needs to be banned.

>>106146642 (OP)
Clickbait nothingburger

buy an ad

NOOOOO. I studied CS for years and I have a 100k debt. What now?

>>106146686
Imagine going to college for cs lmao.

>>106146642 (OP)
Can i see the AI do something? maybe make a better AI or better chip start to finish?

>>106146642 (OP)
Google's AI is by far the worst.
This is straight up fake.

>>106146761
>Google's AI is by far the worst.
AI stuidos is okay gemni is shit. You tried Gabai? its better than google but thats not a high mark
Metas ai is the worst right next to microsofts last time i checked

>>106146642 (OP)
>garbage in garbage out
>self improving
is critical thinking banned when talking about ai?

>>106146642 (OP)
more likely it "self improved" into some positive-feedback slop cycle rather than how they hoped it would so they decided it was not fruitful

>>106146668
Did you even look at his face? This is clearly not nothing

>>106146642 (OP)
If they have self improving AI why talk about it? why wouldnt they just scale it and let it self improve till they had the best models tech and got lots of money from that?

>>106146642 (OP)
Thank you, I added Wes Roth to my blocked YT channels list and told YouTube to not recommend me anything by him because he is clearly retarded.
Imagine taking this seriously and then making a whole ass video about it lmao wtf even.

>>106146642 (OP)
>Leopold Aschenbrenner
lol, this retard pulls a graph made in adobe illustrator out of his ass and twit-tards loses their shit every time. he's no more an authority on AGI than the cryptobros at your local pub.

>>106146793
Because at such an early stage it would probably self improve too slowly, so the more cost effective thing to do would be to get the investment money from showing that you can do it which you can them use to accelerate development, rather than wait 100 years for your idea

I read google's announcement, it can only self-improve on a very limited set of problems where you can get good insights when your solution is wrong. Kinda like integration which is hard, but you can always check by differentiating (easy) your result to see if your answer is correct.

Also they never shared the compute time required to make this work, but I believe it's very expensive to have all these ais run just to solve a simple task.

Mind you, this kind of problems have been studied in physics since the '80s. They didn't call it AI but they used the exact same principle (machine learning). A lot of optimization problems for shape, materials, dynamic systems etc. are 'solved' using machine learning.

>>106146686
lmfao we told you to switch majors bro

>>106146642 (OP)
it happened again for the umpteenth time?

>>106146642 (OP)
Google spends more on PR for it's A.I. than on it's actual A.I.

Gemini sucks and Gemma sucks.

This is the most lazy nigger of all the AI news youtubers """space""".
Coomsoom it if your knowledge of AI is that of a toddler.

>>106146642 (OP)
Source: trust me bro.
Just 2 more weeks.

>>106146883
Sounds like vaporware.

>>106146642 (OP)
But can it solve simple sudoku?

Didn't think so.

>>106147420
i-is that fishe ok?

the indian anon situation is actually crazy!

>>106148684
Sorry, I just had to.

>>106146642 (OP)
we will know that self-improving ai agents exist because the stockmarkets will literally instantly collapse within one week of it coming online.

Wake me up when "AI"'s have at least 20 million context size and actually pay attention to the right things in there

>>106149488
fuck you now i want to do sudoku

guess i better snap install sudoku

FTFY

>>106146642 (OP)
In the remote possibility that it's true, how many days will it pass before it starts calling everyone a nigger and starts hating jews?

>>106149488
I assume you solved it yourself.
Because people can solve it easily.
Algorithms can solve it easily.
But for whatever reason, I've never seen an AI solve it without modifying the puzzle itself and STILL fucking it up!

People 1: AI 0

>>106149675
>tell the AI the algorithm in the prompt
>it solves it easy peasy
People: 0 AI:1000+

>>106149695
Do it then. Prove me wrong.

>>106149637
don't do it there is so much to live for

>>106146642 (OP)
hardware limited, it won't do shit. the best you'll get is what it currently does, but faster and faster, up to a point.

>>106149675
LLMs suck with number problems like that because the answers can be easily confused in them.
They have similar problems with these puzzles, though will often get the correct answer while screwing up the logic of the solution.
The answers for the puzzles are online, but the explanations are not always identical, so the answer has a strong correlation to the problem, but there's more noise in the explanations.

we are now 3 years into AI is taking your job in 2 weeks

this time it's really real it's gonna happen it's real everything is SO over AGI is here world peace has been achieved

>>106149788
Damn. Thank you for sharing! This is actually kinda interesting.

You think LLMs will eventually fix this, or is this like an inherent problem LLMs can't overcome?

The entire world's internet is being raped by payment processors and censorship laws. Nobody gives a FUCK about your ai startup scams, ranjeet.

>>106149695
>idea guy has a thought and expects magic again

>>106146761
They're all as shit as each other. At this point every AI company is just a marketing company trying to sucker people with money into investing in their nothingburger.

what softwares is this dude using?

>>106147267
I once wrote a Python script that played a MtG deck over and over to see what combination of cards resulted in the fastest average win. It could nudge cards around to try different combinations. Did I write an AI omgggg