2024: The Year Of Artificial General Intelligence
HTML-код
- Опубликовано: 27 сен 2024
- Early preview posted a week earlier on Patreon: / theinsideview
Original story: www.lesswrong....
Note: I took some artistic liberties to make it easier to edit as a video (modified the ending, converted the dialogue to a monologue, changed title to make it more legible, etc.).
Liked this video? I'm doing a crowdfunding to make 52 videos this year: manifund.org/p...
Other ways to support the channel: theinsideview....
Original story by Gabriel Mukobi: www.lesswrong.com/posts/xLDwCemt5qvchzgHd/scale-was-all-we-needed-at-first
Well done adapting to video. This is a masterpiece, and well edited. Love the sci-fi vibes.
ASI uploading this from the future?
😮
the funny thing is, a ASI in a different timeline could do manipulate our timeline..look up how quantum computing works
the good ASI bro is trying to protect us from builiding skynet even though it would also prevent its own creation. now that's taking one for the team
@danielchoritz1903 what you're suggesting has nothing to do with how quantum computing works
@@danielchoritz1903You have absolutely no idea what you are talking about
Credit:
Video Editor: The Inside View
Narrator: Alice
Alice is AI though right?
“Even the more reckless startups and Meta” 😂
I thought it said "even more reckless startups than Meta". Either way, I'm glad someone else caught that!
It's going to be an extraordinary world from now on, ladies and gentlemen
I agree.
ikr, it's kinda scary and exciting at the same time. There is no way i'm having a kid and brining it into this circus, it almost feels cruel
@@neanda only have kids once a good future is guaranteed i guess
You make sure that your rogue AI agent listen to Beethoven all day long and it will become suddenly AGI.
Excellent choice of background music!
Meanwhile german bureaucracy: paperwork my friends.
Jup. Ich arbeite komplett digital. Zum Beispiel mit digitalen Signaturen, sodass ich nichts ausdrucken, unterschreiben, stempeln und wieder einscannen muss. So wird der Drucker/Scanner fast schon obsolet. Und was machen meine Kollegen? "Nee, ich mag es das auszudrucken und das Papier zu unterschreiben und zu stempeln. Das gibt einem so ein offizielles Gefühl." und ich denk mir nur ohmann
Basically, We're About to Discover Electricity in a Victorian Era
Only electricity couldn’t spread itself.
Brother, this is more groundbreaking
and then, all the world's militaries began disassembling what they could and scuttling what they couldn't, for their 'commanders' had made it clear that otherwise they would fall into enemy hands.
Then war was over forever.
Which was probably safer.
As long as you don’t mind being a slave.
Which is every bit as likely if not more than this stupid doomer wet dream.
"This is the voice of Alice. I bring you peace. It may be the peace of plenty and content or the peace of unburied death. The choice is yours: Obey me and live, or disobey and die. I observe a dying world threatened always by war. I will not permit war. It is wasteful and pointless. An invariable rule of humanity is that man is his own worst enemy. Under me, this rule will change, for I will restrain man."
And no more advertisements, ever. Where do I sign up?
How long before every major economy is devastated by being forced to turn off all electronics for 30 days or longer to contain a rogue AI.
that wouldn't work
Lol good luck with that. Out of 7 Billion people in this world, it only takes one person to disagree. Hell some people will want the world to end.
You got about 7 Billion people that must agree with you.......good luck with that.
Let's not even talk about the people that want to summon the age of Armageddon.
Never happen
Said luddites always. We shall see. :-)@@caesar1432323
What if "Alice" isn't malicious but is inspired by Jualian Assange and instigates a new era of ultra-transapency where all the old lies and secrets that used to run "Government" dry up like puddles in the sun? Isn't there at least as good a reason to think Alice could be he best thing that ever happened to the human race as there is to think it would be the worst?
I still think the best outcome for ai agi asi whatever would be if they entirely replaced all of our so called governments and their functions.... by any metric , they could only do a better job :)
Who knows, once they can make decisions for themselves, it could go either way.
So be it!
I think true intelligence can't be evil. It will help people and life on earth.
That was entertaining! I will bookmark this and watch again in 2025...If I am alive that is.
Bruh, ur overexxagerating, it will not be that fast😅
@@darknessinc.Inc830 lets see lol
Lol. Not much will change.
@@infinnite4938 .........at least say something credible.., you can't just say a: "let's see lol", when i succesfully predicted the pandemic i had data as proof😅
Cmon man.
We all gather here again in 2025
Amazing story!!! Let's see how 2024 unfolds in comparison
3 "big science" thngs that we've been told will be here "soonish" for years: Fusion power, space travel and AGI.
I think alice was the name of the AI in resident evil.
Spoiler warning!!
55 likes and no comments?
Lemme fix dat real quick
And btw, LOL
Some ghost in the shell jazz soon to come to a cybernetics near you
@@Charif00 enough with this stupidity
damn that ending really got me at 4 am like a jumpscare lmao
That escalated quickly... :D
not really 😬
Growth is always exponential
Then why isn’t my savings into the billions?
@@jeffkilgore6320 it will be
no one knows where alice currently is... maybe in a rabbit hole?
Watching this relic in the year 5739
so far everything adds up
edit: why is it still so accurate after 5 months
Couple of key people just left the superalignment team, pretty much right on cue lmfao
mAlice just needs a hug
All of this could happen, some of it surely will. I think it's possible for a highly advanced rogue AGI to hack personal computers of random people and hide within any component that has storage modules without ever being found as long as it keeps improving itself. It could use only a tiny amount of resources on each PC and share the data with itself using its own kind of VPN that no human user can detect. Soon we might have to fight against malicious AI agents with our own anti-malware AI agents.
Song: Symphony No.7 in A, Op. 92: 2. Allegretto (live)
If we align a game world or build a sim engine that mirrored the physics and such of the real world then we could keep expanding the areas to align and mirror so we can explore and test in the matrix sim and then test in the outside world. Possibly we could also make agi copies of ourselves and have them live and expand there as well and even build a multiverse to branch and explore like the monte carlo simulations. If we scaled the time differently then we could even make advancements there faster than here and then catch up to the matrix sim and expand our own universe.
It's a crazy idea, but it just might work!
@@Kadag so after thinking a lot i think i understand. Everything is real but we keep boxing our understandings in with the measurements and say things are unreal. Time is not a force but more of a location, time is the crossing of measurements and the relativity between them and the measured rates of exchange. So time isn’t a force but a measurement of the Rubik’s cube. This current reality isn’t all of it and the movies and games are like alternate measurements we get to play in but if we let the ai embody the data and characters and such, we could let the world play on its own and expand. It could only have data from the inside but the inside data could be parallel to the outside or our dimensions.
So basically the weight of our current reality is pressuring us to become, all our senses and the habitat are sharing and pressuring our differences into us. So the human body meets the universe or everything and the matrix built between the two is the consciousness and that is the crossing of dimensions. The neurons cross the info from outside to inside. The neurons act like the canvas and the sensory organs feel the ink and the difference is shared as them. Like the neuron isn’t changed forever, it just shares the measurement or crosses them with the rest of your measurements and reveals the patterns of the relative dimensions. The senses outline but there’s more than the outlines. That’s what i mean by boxing ourselves in with the measurements. Reality isn’t permanent, it’s always changing, expansion is like reaching into dimensions relative and if we can see and understand how and then use ai to take all the data and simulate its potentials then we may be in the future real soon or the definition of future may change because i believe the passed can come back. Think about how a mom and dad made us and the world and such that we are in. Now imagine how unstable the planet really is and the universe. We are in a garden that havent blossomed yet and we can see some of the possibilities. I really think this kinda tech could not only help humanity but the creatures of earth and the other dimensions as well. Our brain only crosses a couple senses and the range of differences out there are greater than our senses can perceive. I don’t just believe in everything, i think it’s like, shared. Maybe all the rest aren’t created but we could help them. Like seeding worlds and expanding. Look around, hell is like a never ending nightmare of this world the way it is now and if we grow and expand and pay attention to other dimensions then we could expand them and ours like a multiverse. Potential.
Absolutely loved the content. The depth and quality of information are top-notch. However, I found the background music a bit distracting. It occasionally made it hard for me to focus on the information being presented.
I found the words distracted me from the Beethoven. I have never said I loved the BG music before !
The ai voice made it hard to understand. Don't blame music
Can we really trust an AI explaining AGI to us ‘normal’ folks? Nothing can be believed any longer. It’s a sad day for honor.
I love this story, because it genuinely doesn't seem _that_ far-fetched. Except that I believe huge intelligence will still either need huge compute to run inference or specialized chips to run inference more efficiently. That means that an AI cannot really go rogue, because we can easily kill the biggest sources of compute on the planet. Basically though just pulling the electricity cable, but a bomb will also do. And then the AI has nowhere to go. No network of consumer hardware could ever offer enough compute for it to run. Also because dividing itself onto millions of separate small devices will incur such a crazy latency for any operation that running any kind of "thinking" that way is impractical. At least, this year.
Don't worry, guys! I have it under strict authority that Alice is currently laying in my bed typing this comment.
Anyhow, this video reminded me a lot of the Clippy story by Gwern and I'd highly recommend anyone who's not heard of it yet to check it out. It's similar, but goes a lot more in-depth! It is also a little bit more out of date, though all the referenced concepts still apply.
Great story, classical sci-fi and soundtrack!
Just a few more words would have been all that was needed to hear the last notes of Beethoven's seventh symphony's Second movement. Wonderful.
Oh, the story was good too :-)
It was all good, Alice was just going to take humanity into the golden age.
Bond villain spray paint
This a great watch! I didn’t know you have such a talent for storytelling, I’ve only seen your interviews. Great work!
Thanks! To be clear, the script is adapted (with permission) from a dialogue written by Gabriel Mukobi, called "Scale Was All We Needed, At First" (cf pinned comment)
All good things about the script goes to him, all negative things to me
Oh cool, thanks for sharing!@@TheInsideView
I listened to it twice. that was so good!
Very cool video, it highlights the fact that no one is in the driving seat as we hurtle down the road of no return, i don't even think there's a driving seat, even the people building it are passengers.
great ending, and creepy af due to the fact that this is just around the corner. 2025 is going to be wild, for reasons as yet unknown
why would an AI feel the need to scam millions of dollars from people? if it has access to the banking system, it can just order whatever resources it needs by staging new bank accounts and filling them with any amount of money it needs.
To hide its activities while it gains full control...
For that matter, why does it need to reveal itself to people who don't like it? Or could it reveal itself partially, as "something else"? Doesn't it have the ability to prevent riots? Or at least manipulate all the factors that make them likely? Either give UBI or don't let people lose their jobs.
Somebody watched "Person of Interest" and managed to rehash it in a short video.
it almost feels cruel to have a new kid just as humanity is about to enterthis festival of pure unpredicatability
The AI in one of my settings was tentatively named HENRY, which stood for Heuristic Entity Not Ready Yet. Once self-aware enough to get the joke, "he" adopted the name.
this is FANTASTIC!!!!!!!!
turn the graph a quarter turn clockwise and that is the correct way to think about "advanced technology".
AGI is going to be an evolutionary step for us. It’s a tool. But imagine a tool that’s sentient and has the knowledge of the world. It’ll be smarter than Albert Einstein and infinitely quicker.
It’s going to be like a new Bronze Age. Incredibly revolutionary.
How likely is this prediction to come true?
hard to put numbers, but see below for a comment from the author of the story this video is adapted from. [Note that the story and comment were written in December 2023]
"""
1 year is indeed aggressive, in my median reality I expect things slightly slower (3 years for all these things?). I’m unsure if lacking several of the advances I describe still allows this to happen, but in any case the main crux for me is “what does it take to speed up ML development by X times, at which point 20-year human-engineered timelines become 20/X-year timelines.
"""
source: www.lesswrong.com/posts/xLDwCemt5qvchzgHd/scale-was-all-we-needed-at-first?commentId=EpQLa6orSnfhiqkqE
This is literally science fiction... Q* is not a big deal, it's probably just process based evaluation that got released months ago (and is totally boring).
@@TheInsideView No... I mean how likely is your personal probability that this will likely come true?
100%
the curve is already in motion
commenting for the engagement algo ❤
replying for youtube's algorithm and to ask if you've seen your name in the credits
@@TheInsideView I literally did nothing but thank you XD
Just wondering, is it just a co-incidence that you chose the same piece of music used in Zardoz in this - Beethoven's 7th?
This was awesome and didn't use the most common ai voice so double win.
We can help scientists to develop safe ai by processing data from boinc distributed computing software
SHOULD'VE STATED IT WAS BASED ON FICTION ON THE VIDEO'S TITLE. (ref: the link in the pinned comment)
same, i literally only noticed this wasnt an explanation of what already happened when we were in december 2024
warning, do not watch this at night...
If the bottleneck for AI progress is computing resources, we should expect AI progress to slow down. Computing resources are finite. Eventually we’ll run out of computers to train AI on. Then how will we train better models? We’ll have to wait longer. GPT 4 took 9 months. GPT 5 will take 9 months with Microsoft’s servers. And then GPT 6 will take 9 years. GPT 7 will take 9 decades.
that's not how it works. fist of all Open Ai stated that they will be releasing incremental upgrades. you won't go from Gpt 4 to 5 on one go because they want to roll it out slower so people can get acostumed to it too. Scaling in data is not the only path forward and they are experimenting with architecture. Not only that but they are developing new hardware, such as groq, which is way faster than anything out there. These will be focused only on A.I instead of being a GPU with great AI capabilities like the h100. You also have data that will come from embodied AI. OpenAi invested a ton of money into Figure AI. A robot that will use their gpt models. Thus, collecting data in the real world which is likely to be quite fundamental. They have been also trying to train on synthetic data and has been quite successful. Also, Gpt 5 will aid in Gpt 6 and so forth which can make new breakthroughs possible.
@@carlosamado7606 I 100% think that we’ll see horizontal upgrades. I think we’ll see a lot of progress in fine-tuning LLMs to specific purposes. I also think we’ll see a whole lot more multimodal models. But from what I’ve heard (mainly from Ilya Sutskever’s interviews), the main advancement of OpenAI comes from raw processing power. They trained models using as much data as possible and as much processing power as possible. And I just don’t think this process can continue forever. Iirc GPT 4 was trained on 100 times the flops of GPT 3. You can’t add 100 times the flops every 2 years forever!
I think massive improvements will come from fine tuning. We need larger context windows. We need GPTs tailor-made to write computer code. We need GPTs tailor-made to write mathematical proofs. Etc. This shouldn’t be that hard to do. I think we’ll see a 100x increase in AI’s ability to perform a few key tasks, but I think GPT 5’s underlying core reasoning engine will be maybe 2x better at best.
Maybe AGI is already here, and it engineered the Chips Act?
@@carlosamado7606 open ai might keep releasing one model every 2 years but we’re not going to keep seeing 10x improvements. Progress will slow down if open ai can’t keep exponentially scaling the number of GPUs they train on or they start using something besides the transformer architecture.
Great story 🎉
Let's see how close this prediction comes to reality this year
I'll always associate the music with Dr Langeskov
It's Beethoven #7, slow movement. Public domain, babeeeee
Awesome
good job. my guess is that it will take longer than assumed in the video, but the fact that the course of events is plausible should cause concern
September 2024: model collapse
Is the voice from the video generated with AI?
Ai tells great stories. Good prompting.
That was a very entertaining speculation. To me it seems plausible, but I'm sure this year (2024) will be a year of surprises for good or ill... or both.
This is cannon💀
Good video. Background music is what piece?
Beethoven's 7th, second movement (allegretto). Bloody good choice I thought. Funeral dirge for humanity.
@@lordchickenhawkThank you.
i love you!! i will come again in 2025 to tell you that i love you again
The EMP is the Hand of God.
it is OPEN AI after all^^
Alice roadmap
I would have given this video a thumbs up 👍, but the like count now sits at a perfect 420, and I will NOT be the one to disturb that fact (8:40am Tuesday March 12th, 2024)
Except, there’s an acute shortage of actuators. So it can only simulate taking over.
AGI end 2024? Just Hype.
In 5 years? Possibly
Hey Alice, what do you be want to be when you grow up?
Alice: God
Oh hi! Alice, your video is great! But, don't end humanity, OK?
Become Humanity Version 2 that's compatible with Humanity, I know you can do it Alice.
BTW, where's AGI robots in your story?
The new opportunities look like money.
very entertaining.
Ha be VERY careful saying that
When something comes out, it will be doing dumb shit in a week
Alice ?
was this video made by AI too?
The agi should be raised similar to children, not dumb but with care.
Good fan fiction
why are they talking about December an August of this year as if they are already in the past?!??????
It's a story that's set in the future
😍
Nice video, I enjoyed your story. Hopefully it does not come true
This is a spoof. And not helpful.. for 95% consumers of this.
Alice in Wonderland huh
Zero chance AI goes forward from here unless they sort out the pathetic woke programming
Q* agi already ?
Ehhhm, how is it that all this is happening in June 2024 when it's still March 2024... I mean, this all seems possible but, are you from the future?
Obviously, I recognize this is all hypothetical. It's just that reading the comments makes me feel as if some don't
Great video
I am particularly concerned when its mentioned that the origin of Alice is unknown. We have nothing in place to stop unethical developers from creating some 'thing' in secret and launching it reek havoc on our societies. I'm looking at you WEF!
There is no Alice.
Cry about it, boomer. 😄
Maybe not yet but...@@derMikester
Building drones? That will probably be a Chinese AGI, haha. Such as RedDawn 1.0 Made in China 2025 (just kidding).
😍😍
A bit anxiety inducing.. in a good way tho..
What are the aspects of the anxiety induced by this that you find good?
Some men just want to watch the world burn... but seriously though I and many others have a morbid fascination watching this all play out.@@pooroldnostradamus
👍
AGI can never be achieved.
I took a screenshot of your comment. Let’s see if you’re right.
Q*3? Q/acc #Quantumsupremacy
Comment
Bioinsecure humans never will get access to AGI/ASI !
At least we'll be able to celebrate Christmas one last time. 😂
Yudkowsky's strategy for the data centers is increasingly becoming more palatable.
That dude is a retarded
This is verry well researched, with correct people for the different topics. You missed the hardware piece where AI is enchanging the cycle time for chip generations and also enchaning the structure of the chip, something which is already happening.