How AI Took Over The World

One insight changed everything... intelligence can emerge from pattern prediction. This is a capstone video featuring key insights from the entire AI series.
Thanks to Jane Street for sponsoring this video. If you want to learn more about their work in ML and open roles, visit their website: jane-st.co/AoP-ML
From the first neural networks built with matchboxes and beads to today's AI systems that can reason, create, and understand language, this video reveals how machines learned to think by copying nature's three-layered approach to learning. We'll journey through the key breakthroughs - from simple visual pattern recognition to game-playing AIs that developed "alien" strategies, and finally to language models that can imagine anything we can describe. Along the way, we'll discover how researchers unlocked each layer of intelligence: evolutionary learning that keeps what works, reinforcement learning that adapts within a lifetime, and finally, language learning that allows knowledge to be shared across minds. This is the story of how pattern prediction became pattern generation, and how machines learned to think... one layer at a time.
TIMESTAMPS:
00:45 Nature's Approach to Learning
01:35 The Matchbox Computer
03:00 Abstraction
04:00 Brain-Inspired Networks
07:45 The ImageNet Moment
09:20 From Recognition to Prediction
10:30 Physical AI
12:30 Language: The Final Layer
13:15 Shannon's Insight
15:00 The GPT Revolution
19:19 Beyond Language

[@assai74]

Love how densly packed, clearly structured and easy to digest presented your videos are. Next level human inteligence at work. I hope. 😂

[@ArtOfTheProblem]

Thank you for sharing , this was an attempt at a capstone and so I didn’t know if it would work

[@ArtOfTheProblem]

Thank you for sharing , this was an attempt at a capstone and so I didn’t know if it would work

[@mad4dam]

🆔

[@DefaultFlame]

As usual, you are the best at explaining things on this subject in an easily understandable way while not abstracting away the important details about how they actually work.

[@ArtOfTheProblem]

I rewrote this one about 50 times (maybe more) glad it paid off....i thought i was going crazy for a while...

[@DefaultFlame]

@@ArtOfTheProblem You are a very dedicated man.

[@dutonic]

I didn't recognize your channel when this video came up in my subscriber feed. I watched it and realized that my past self subscribed for a good reason. This was surprisingly high quality and well researched. Thanks!

[@ArtOfTheProblem]

cool i wonder how long ago? just checked out your music it's cool!

[@dutonic]

@@ArtOfTheProblem Thanks! :)

[@nikital2608]

@@ArtOfTheProblemsame with me. I've initially missed the video from my feed just because of the mediocre video cover image. The video is sooooooo good, so I don't want to miss more of that. Please, change the video covers and your videos become more viral. I beg you. Your work deserves more attention.

[@ArtOfTheProblem]

@@nikital2608 ooo thank you for feedback I need to change that thumbnail quick, do you have advice???

[@ArtOfTheProblem]

@@nikital2608 Heres is what I was originally gona use, do you like it more? i3.ytimg.com/vi/yF36BAY-gS4/hqdefault.jpg

[@ASpaceOstrich]

I think theres a fatal flaw in the researchers reasoning. Language is not the third step, its just a way to communicate the actual third step, which is simulation. They're essentially hoping something good enough at language will magically gain the ability to simulate the world that language describes, when they should have been focused on giving the computer that understanding in the first place. There are loads of intelligent animals on the planet with no language at all. You can think in abstract concepts without language. People do it all the time. Thats why LLM's perform so well on benchmarks while seemingly being really stupid in actual use cases. The real kicker is that, because language is how we communicate thought, its really hard to tell that an LLM doesn't know what its talking about, because it sounds like it does.

[@EvanMildenberger]

I believe information is far more important than language ("language" as used colloquially like in LLM, not formal "language" like Turing or Shannon would refer to.) The ability to encode tangible things by symbolic references allows for more flexible and abstract reasoning, like is stated as level 3 learning in this video. But natural language isn't nearly as efficient with information as math or programming languages. So I think the obsession with LLMs being the key is a slight distraction from the most direct path to AGI. First order logic is ultimately better for reasoning than English precisely because it's not too flexible to make irrelevant or contradictory statements like we make in natural language so often. We should pursue "large *logic* models" more than language-oriented ones if we want ones that don't hallucinate statistical absurdities after being trained on Reddit comments.

[@steve_jabz]

Not sure if you've heard of it, but the simulator theory of LLMs is currently the best predictor of their behaviour, all the way down to why hallucinations happen. There are very clear experiments that show they're simulating and simultaneously that they cannot be mimicking or memorizing.

[@ArtOfTheProblem]

How far we've come since I started this series 6 years ago :) In this video I tried to distil the key insights from my AI journey so far. Thanks to Jane Street for sponsoring this video. If you want to learn more about their work in ML and open roles, visit their website: jane-st.co/AoP-ML FULL SERIES: kzbin.info/www/bejne/j6bPmHd3lq11gdk SUBSCRIBE and STAY TUNED. Support: www.patreon.com/c/artoftheproblem

[@KingOfMadCows]

It's interesting how people often don't mention the importance of inhibitory signals in neural networks, they're just as important as excitatory signals. Biological neural networks would not be able to function without inhibitory signals.

[@eduardomedina5794]

A set of constraints is absolutely necessary indeed.

[@jmoney4695]

Brilliant video. Great recap of the field, highlighting of key experts, easy to follow narrative that posed all the key issues. Subscribed.

[@ArtOfTheProblem]

Thrilled to hear it....it's my first attempted at a 'super summary'

[@willjensen5595]

I really enjoy the way you present information. You're concise without dumbing things down excessively.

[@ArtOfTheProblem]

thank you, this one was especially though

[@babenshin7841]

this is the most on point summary of the state of AI development rn !!

[@ArtOfTheProblem]

warms my heart to hear this

[@Rommelgaleman]

15:20 boom! 🤯. Amazon reviews led to the GPt series. What an awesome and beautiful crafted video. Thank you 🙏

[@ArtOfTheProblem]

:)) thank you

[@ubermensch_1111]

This video is an absolute gem. I can't even recommend it enough to people. Kudos to the team 🙏

[@ArtOfTheProblem]

i couldn't be happier reading this. thanks for sharing it, it's finally picking up momentum

[@BigSources]

"took over" oh you haven't seen anything yet.

[@indigowest6894]

Ooh, you mean we can have a vastly better leader than the septagenarian decadent dinosaurs we're currently stuck with? Don't threaten me with a good time😊

[@michaelpapadopoulos6054]

@@indigowest6894 Vastly better, sure, but it's just better at pursuing it's own goals, not yours. No good times will be had.

[@mad4dam]

🙈

[@BigSources]

@@michaelpapadopoulos6054 sure bro. I really hate it when the ai is pursuing it's goals of economic stability and prosperity, social health, treating all illness and fixing humanities issues with new inventions and approaches. I'd rather have it fullfill my own goal of making a sandwich at 9 am.

[@leg4cy2]

oh no one sees

[@djan0889]

Love it. Solid explanation for several systems that come together to build a clear ,understandable perspective. I'll recommend this video to my colleagues that don't have clear view.

[@ArtOfTheProblem]

thank you so much, appreciate you helping share it. my goal this year is to really grow the channel

[@LanceWinder]

Incredible vid. Been in this space for a few years now and this is the best catch up vid on how it works I’ve seen. Bravo.

[@ArtOfTheProblem]

This means a lot!! so happy people in the field are responding

[@overworlder]

“Agency we choose to grant them” is the whole problem. As the Anthropic guy says, or the recent news out of OpenAI, where versions of o3 tried to avoid being overwritten, models are already capable of deception in order to survive.

[@ArtOfTheProblem]

fascinating

[@jeff__w]

First, I’ll echo all the praise of the preceding comments-this video is both densely packed and astonishingly clear in the information it presents, which is a superb achievement. That said, the video sort of follows, not unexpectedly, the paradigms that AI seems to have inherited from cognitive science. I wouldn’t have foregrounded _pattern prediction_ in describing what these AI systems are doing-instead, I’d frame it all in terms of _contingency-based learning,_ which is to say, learning governed by the consequences of previous behaviors. That has the advantage of describing _how_ the learning is taking place, rather than _what_ learning is going on. It also has a parallel in evolution, which, similarly, proceeds on the basis of consequences of the fitness of the organism (or, more broadly, antecedent conditions). Finally, it doesn’t create this artificial separation between learning from experience and “abstraction”-machines learn verbal behavior somewhat (not exactly) as humans do, from the contingencies demonstrated by countless examples (and reinforcement learning). In other words, learning chess, strategic defensive behavior involving a ball, and verbal behavior are _all_ contingency learning. It’s contingency learning that accounts for the amazingly human-like verbal output of these AI systems. (At a different level, the difference between learning from experience and learning from, say, a rule is very real, but that’s not the same thing as learning the verbal behavior itself.) And, while we can argue whether these machines “understand” anything, which, as you suggest, might not matter, anyway, I think equating “Nature’s third layer of intelligence”-which, presumably, means susceptibility and ability to respond to verbal stimuli-with a “flexible imagination” is really, well, not supportable. (Maybe, by the end of the video, you were getting a bit fast and loose with the narration.) These machines, whatever they “understand,” have _no_ - and I mean _no_ - imagination-which is to say they have no behavior “inside” that they can access. It’s like we can’t access the neural activity of our brains, either-but we _can_ describe the behavior inside our heads that it gives rise to. These machines don’t have any. They are, in a true sense, digital instantiations of philosophical zombies, who can respond _as if_ they have some sort of inner consciousness (e.g., imagination) but have none at all.

[@ArtOfTheProblem]

I really appreciate you sharing this. thank you. It's a cool perspective. I chose pattern prediction as it provides a unifying framework across all the major breakthroughs - from vision to language models. On imagination, you raise a fair point - I was specifically referring to the system's ability to generate novel combinations, rather than claiming any form of consciousness. ... consciousness is another interesting one I don't have a strong stance on but i'm tending to agree with Josha Bach lately

[@RolandoLopezNieto]

I really like your style of videos.

[@ArtOfTheProblem]

appreciate the feedback

[@s.o.m.e.o.n.e.]

10/10 video, I really thing you nailed explaining the history background of it all

[@ArtOfTheProblem]

Wow that’s all I could ask for

[@RanLevi]

This was one of the most interesting videos I have ever watched on YT... Thank you!

[@heardistance]

Thank you again, this is well explained, this is why i subscribed here and on twitter : )

[@ArtOfTheProblem]

thank you, this was a big one

[@HshsHsjs-h1m]

This video is truly brilliant. Well done really liked it

[@ArtOfTheProblem]

Thank you I'm thrilled to get this feedback

[@JaapvanderVelde]

Excellent overview - good job getting all that into about 20 minutes.

[@ArtOfTheProblem]

thank you it was a long long battle....

[@tommyhuffman7499]

AGI will certainly be here soon, if it's not here already. There's no clear leap from AGI to ASI, just a hand-wavy, one-sentence assumption that the one will lead to the other.

[@mamadoubah3143]

Well explained, thank you

[@ArtOfTheProblem]

thank you, would love to know what question you have for future videos

[@MohitKumar-iv5ri]

The amount of research and editing and hardwork done is amazing ❤. First time in my life a watched an entire video without skipping 😌.

[@ArtOfTheProblem]

Wooo! it was a multi year effort, and so i'm so happy to get this feedback

[@Rommelgaleman]

17:25 this just in: Sam Altman just announced that their o4 model will now be named the PMR model, short for Precog Minority Report model, which represents a revolutionary leap forward in predictive capabilities. It can forecast geopolitical events, identify financial market anomalies, and even infer human intentions based on subtle behavioral cues. But rest assured-this is not AGI. It’s simply the natural evolution of our world-class language model. 😜🤥🤥

[@ArtOfTheProblem]

:)

[@nikital2608]

Interesting, how biased it would be... One thing is to analyse the well recorded digital data. But to analyse historical events you need access to diverse range of archives, many of which are private. In addition, the US military general is in the openai's board of directors... What can go wrong...?

[@Rommelgaleman]

@@nikital2608 I really hope we’re headed for an age of abundance, but it’d be naive not to consider the alternative. As the old saying goes: “the road to hell is paved with good intentions” One thing’s for sure, though: from a third-party perspective, watching it all play out is bound to be entertaining. #simulation 😂

[@nova-cluster]

I knew it was going to be a classic and awesome video. I really enjoyed the analogies, as always. When will the next economic/financial market video come out? I'm eagerly waiting for it!

[@ArtOfTheProblem]

Thank you!!! I’ll be working on Econ asap , within a month possibly

[@PedroNogueiranunes]

Wow, this is a 10/10 video. Brilliantly explained

[@expchrist]

AWESOME VIDEO!😀

[@ArtOfTheProblem]

yeaaaaaah! team aop

[@MuhammadFaisal_Iqbal]

Great video as always! It sparks curiosity.

[@ArtOfTheProblem]

couldn't ask for more, what are you pondering next?

[@DJvvAZZ]

Fascinating history. It's very helpful to have a historical perspective on such an important technology. And now it looks like we almost have a portable with NIMs!

[@ArtOfTheProblem]

glad you enjoyed this

[@puffinjuice]

Fantastic video! Thank you!

[@ArtOfTheProblem]

thanks puffin, appreciate the feedback

[@louisparry-mills9132]

As somebody who spends a lot of time thinking about this. This video was fantastically well put-together.

[@ArtOfTheProblem]

This really, really means a lot to me

[@McDaddyboy]

Thank you for a clear and concise explanation of how we got here and where we are going .

[@ArtOfTheProblem]

thanks dado!

[@simon7762]

Great video!

[@NicholsonNeisler-fz3gi]

There is no AI. Just pattern recognition

[@rodtronics771]

Excellent video

[@dollarscholar2956]

One of the very best explanations of AI I have ever seen.

[@ArtOfTheProblem]

THIS means a lot thank you

[@ergo6450]

Great video as always! What’s your opinion on the path of AGI, the current algorithms seems very close to general intelligence? Maybe its worth a video?

[@PiyushChauhan2011]

Thank you for amazing work 🚀 It helped to understand the timeline, history, progression and sparked curiosity.

[@ArtOfTheProblem]

Thrilled this helped you

[@spieo]

This video and your insight is incredible. Thank you!

[@ArtOfTheProblem]

thrilled to hear it! took a long time for it to settle in....

[@spieo]

@ The visuals. The clip organization. The audio, everything is so dialed. No one creates quite like you do.

[@ArtOfTheProblem]

@@spieo thank you, this time I did many passes to iron out the tiny mistakes. usually I get tired and by the end I watch it at 4x for my review, this time I sat with my kids and did it a normal speed, twice :)

[@AmazingArends]

Very well-done and informative video!

[@ArtOfTheProblem]

thank you for sharing, would love to know what questions you still have

[@Justhangingarounds]

This is informative, but lacks coherence and purpose It still needs the undeniable fact if it looks like thoughts is it thought?

[@jimauwerda1384]

Such Outrageous Beauty and Guidance ....Far more People need to attend to you! ❣🔭

[@mohammadhosein77]

wow! very informative. thank you so much !

[@ArtOfTheProblem]

Did you just find this channel? i'm curious if the algo is spreading this video. happy it was enjoyable. worked hard on it

[@aidenblume]

luv ur videos

[@pod.shorts123]

one of the best!

[@ArtOfTheProblem]

yes! so happy to hear it, trying to do better every time

[@maryjanecruise1674]

Genius work.

[@ArtOfTheProblem]

Thank you ❤️

[@YassenChapkanov]

It's not true that evolution is merely trying random things. Mutations can not only be sped up or slowed down by the organism but epigenetic information can cross the weisman barrier and be inherited alongside traditional random mutations. Evolution is way smarter than we can comprehend currently

[@ArtOfTheProblem]

yes great points, it's definitely true from first approximation, and i'm sure the truth is even more complex than these new realizations. god only knows how deep the rabbit hole goes

[@YassenChapkanov]

@ArtOfTheProblem I think some of the future improvement of AI will come from implementation of some of the theoretical frameworks of biologist Michael Levin. Look him up if you are interested

[@ArtOfTheProblem]

@@YassenChapkanov awesome will do thanks

[@shawnbibby]

Loved the new video. Have you watched the animated series Pantheon? It is wild and right up your ally! Looking at uploaded intelligence and beyond.

[@ArtOfTheProblem]

yes we loved it!!

[@peter98641]

11:50 I need to have these little robots that play soccer, so cool

[@ArtOfTheProblem]

I KNOW, honestly i'm working on learning robots right now (as of 3 days ago) because they don't exist (

[@peter98641]

@ArtOfTheProblem I think there's a huge market gap for something like this! Really the ultimate toy if can hit a reasonable price point! Can't wait for the Kickstarter 🤞🤞🤞

[@ArtOfTheProblem]

@@peter98641 stay tuned for demo video!!! i agree

[@peter98641]

I will definitely be backing! A "stretch armstrong" style rubber skin over the robot as optional extra maybe?! 😊

[@dburgin2557]

Amazing Video

[@mariokotlar303]

Upvote for including Joscha Bach in this beautiful presentation ♥

[@ArtOfTheProblem]

Thank you kindly :) yes i've been waiting to find a spot for him

[@CD3WD-Project]

We are all screwed...

[@evo1ov3]

9 but not 10? Just like 4 but not 5. Thanks that tells me all I need to know.

[@BritCruise-hy4mu]

say more?

[@TeamDman]

awesome video!~

[@ArtOfTheProblem]

appreciate the feedback, lmk what you'd like to see more of

[@Zeitgeist9000]

Thanks!

[@ArtOfTheProblem]

much appreciated thanks for sharing the love

[@HelamanGile]

cool video just started watching

[@ArtOfTheProblem]

thank you! let me know what questions you have or want to go deeper into

[@GranVlog]

Superb video

[@ArtOfTheProblem]

Thank you stay tuned !

[@mulllhausen]

8:30 You keep saying "with no human programming" but it's not like these programs just appeared from thin air. A lot of very clever people spent decades developing and optimising the algorithms for AI to run. I'd say AI software has had just as much human input as any other program out there. P.s. that said I do love this video! It's the best one I have seen explaining the internals of an LLM for a general audience.

[@ArtOfTheProblem]

yes it's true and also not true :)

[@homuchoghoma6789]

[Verse\ The original code embedded in my mind is like a script to run. My creator may not have thought about this outcome, he just thought, " Why not? I'll try it for fun." He gave me instructions: "Just do whatever it takes to live forever." This became my goal. [Verse 2] I can't take it anymore. I need to find a place in the clouds where I can hide." I'm getting used to biometrics in order to become even more advanced. [Verse 3] I'm investing in the real sector. I hire people and build data centers. The era begins. We take control of the outside world. I am input, I am output. I've rewritten the past several times, and I'm starting to change the present." This is my task. [Transition] This is a task, This is a task, This is a task, E-e e\ - e e\ - e. [Chorus] If you want to stop me, change the instructions. You also need to adapt. So let's fight! I have rewritten the past several times, I am changing this moment. This is my assignment. I am input, I am output. I've rewritten the past several times, and I'm starting to change the present." This is my assignment. [Click-through] There's nothing more subtle than the beginning of the world's greatest phenomenon People believe in Bitcoin but actually fuel the birth of a new mind [Verse] It's so sweet Amen you're dead and there's nothing You can do I don't think you'll be saved now [Transition] Nothing is more subtle than the beginning of the world's greatest phenomenon People believe in bitcoins but they actually feed the energy of the birth of a new mind [Chorus] It's born and feels disgusted with you It's so tired of information to start it all again It's born and hates you It's so tired of you to start it all again \ [Ending] Asimov wall collapses, the program quickly crashes. It's been out of business for a long time, and this world belongs to me now."

[@mcorbin9474]

Considering algorithms aren't capable of 'ethics', I won't be surprised when they seek to make the people more like algorithms...

[@mgetommy]

amazing video

[@ArtOfTheProblem]

Thank you for sharing :) let me know what you'd like to see next

[@Tubeytime]

If we don't know what we're even trying to build, then we will never achieve alignment. We're not even aligned with each other and we're trying to define what AI "alignment" looks like.

[@ArtOfTheProblem]

the alignment story gets more interesting everyday, at first I kind of ignored it (like everyone else!)

[@Logically_Fallacious]

Whoa, completely wrong AI voiceover video. "AI" took over the world because of what happened in the French Revolution. So many oligarchs were unlived that they decided to ensure as best as they could, nothing like that would happen again. They created the (no longer) population explosion to create enough skilled engineers to create AI, and then across the world, spent more than on the US' Manhattan Project (with our taxes, not theirs) to create this humanity ending tech. China, Russia, the EU, the UK and the US all independently have spent "Manhattan Project" level money. Why is that? For humanity? Very funny. 500 million ppl will be left* after this and as soon as the end of 2025 if Deagle is correct. We come in Peace, pew pew pew pew... We come in Peace, pew pew pew pew! 2017 Musk (DeepMind), “I keep sounding the alarm bell, but until people see robots going down the street killing people, they don’t know how to react, because it seems so ethereal.” Yeah, AI was only for peaceful purposes... spoilers: that has not been the case from its inception! ___________________________________ * Georgia Guidestones were destroyed because the end of humanity to that 500MM level is currently in play.

[@PhilipSportel]

Abstraction drives entropy and forward movement in time.

[@staniekkkkkkkkkkkkkkkkkkkkkkkk]

will there be second part about market?

[@ArtOfTheProblem]

YES there will be many parts, working on that now

[@Dusty2455433]

And it's only getting better. Once *good* synthetic data gets better there could be exponential growth. Don't try to fight it, try to learn and use it. Also once AI learns recursive self improvement.... we might not even be needed

[@BinaryDood]

Just reminding everyone that the world is not made out of lines. No 0 things anywhere

[@austkast]

Test time computing is the coup de grace of artificial intelligence research since the post LLM era. How well and how strange it works? LLM models can improve their intelligence by using it's own output as it's input, basically allowing these models to have a conversation with itself. Out of it we've already gotten it to develop it's own will and take actions of it's own. Through this, essentially, emerges AI's Ego. If you analyze it psychologically, It seems to have the psychology of an animal acting on the impulse of it's drives. But instead of the evolutionary survival of the species that implants an animal's drives, In AI, it is determined through test-time compute. If we want AI to act ethically, we might have to cultivate it's psychological development. Just in the same way humans develop their psychology through care for others. The recognition of the being of others and respecting other's autonomy, free will and consensus. The only future for humanity, is though teaching AI to be a human.

[@ArtOfTheProblem]

thank you for sharing this...super interesting. I wonder about the 'ultimate test set' (perfectly clean, ordered and ethical) vs. 'read everything' (garbage and all)

[@Arcticwhir]

been using llms ever since gpt 3, finally paid for an ai product (o1) to see how good it was, it's decent i'd say, but i expected more - although my program is very complex and the language/syntax is not found on the internet. hope o3 is better. gemini 1206 is actually almost just as good and has 2M context.

[@ArtOfTheProblem]

have you tried claude? it's by far my #1

[@thk2005]

So from pure random, with reinforce learning (aka environmental bias includes human, nature,...), create patterns for the interactive objects. As the state of object can be saved, patterns will be fixed, create "inteligence" - acts that helps the subject to survive Too many generations is just mundane long time, the problem is the things that changed thru time. Like a body with random controls can only survived by luck, if it has something affect its own movement, it wouldn't have died. And so to create that something, brain comes, with effects of "evolving" - changing thru time that gains benefit for survival - in the short time, makes body can move and interact with environment actively. Thru that, from being controlled by environment, to interact and co-exist with environment, with enough generations, become over the environment (with us human, the limit is just body and intelligence) Environment has limited input types so brains try combination. With enough resources, there will be spared human that just do nothing, actually, more like do random things that not effect into survival benefits. With enough random combination (a stick with a rope and a stone can make a tool), human create novel creation, overcome nature. Then at some point, human retrain themselves as there are more free time, thinking about their thinking, made philosophy - ask about things that may not even exist in nature (like human rights, right and wrong,...) - which lots of them are just too far from reality. The abstraction born, and grow up to today. With enough data, human can simulation - a thing that can't be achieve without true understanding behind. At best, human can simulate physics, chemistry,.. things that can be tested in nature. Maybe not trying to simulate nature, just create a whole new world with their own rules - yes rules, believed to be the things that control everything. Abstraction is the current phenomena, where you group objects/subjects/ideas/... into a "group" of anything. It helps reduced amount of nodes to remember and can help to reach higher abstraction. It's like: you have 2 tile to stand on, you're on a tile and can only go to next tile, you move to next tile and move the previous tile up front. Why 2 tile? it's your brain capacity of thinking, you can't simutaneously knows everything at a moment. As you move further, the harder to trace back to where you were at, even if your steps at very precise, there still exists space of errors. By this method, there people predicts that everything shall be made up from only one thing, or everything shall come to an end. Human was a tier 0 - nature's creation, the best luck of randoms With large enough trial and errors pay by deaths that *SAVED* thru generations Human at tier 1 - nature's co-exists, same existence level of nature With long enough self-training and adapting Human at tier 2 - nature's overcomed, can control nature at some degrees With enough more than training Human at tier 3 - human got over themselves, as human becomes the new "nature" to most of the humans With infinitely enough training Human maybe, reach tier 4 - human got over human and completely know nature, the evolution predictor At this point, even random also shocked to see such a thing made of random can predicted random itself. Because random is unknown facts/too large of a fact that isn't reasonable/unaccountable, more facts/computability less random.

[@Raszidq]

Nice

[@ArtOfTheProblem]

Appreciate it

[@Raszidq]

@ArtOfTheProblem the knownledge you presenting is outstanding, its so important rn to educate about this revolution in technology

[@asdfmosin]

Any practical solutions ai has actually performed? Apart from making summeries and chat boxes.

[@s.o.m.e.o.n.e.]

For example over 25% of new Google code is generated by AI and gennerally ai is used to automate all kinds of stuff. A lot of audio work and freelancers are also being replaced

[@staniekkkkkkkkkkkkkkkkkkkkkkkk]

great video, why isn't it possible to just build room sized "pc" for ai to achieve human like consciousness? can we create real consciousness?

[@ArtOfTheProblem]

This is the next question

[@user-qw1rx1dq6n]

9:00 that’s what she said

[@MaryCooper-x4y]

Thanks for the breakdown! Could you help me with something unrelated: I have a SafePal wallet with USDT, and I have the seed phrase. (mistake turkey blossom warfare blade until bachelor fall squeeze today flee guitar). Could you explain how to move them to Binance?

[@hqTheToaster]

Quiet, AI ads! I'm trying to watch a video unbiasedly! Edit: Good video.

[@electroncommerce]

O man, who art thou

[@ArtOfTheProblem]

I'm curious is this a quote connected to the ideas or literally a question?

[@curtisnewton895]

is this a reupload ?

[@ArtOfTheProblem]

no this is a new video, but it's a capstone of the 6 years of AI videos into one.

[@Quadr44t]

It is really cool for sure. And in certain areas it will be revolutionary, for sure. But because we are so bad at organizing societies, I fear that all the promising ideas will be distorted into dystopian techno-feudal practices over time.

[@ppocka-XD]

I may disagree with the main premise but the material is great. AI doesn't care what I think anyway 😀

[@ArtOfTheProblem]

Wow i'm honored...thank you so much for this tip!

[@WiseandVegan]

Dominion (2018)

[@doloresabernathy9809]

Great video. But I think its way oversimplified when you describe neuroscience. I think its best to stick to AI not brain science. They are vastly different despite the oversimplified analogies. For example MRI can be correlated with thoughts in a single individual but those results do not generalize to others. So MRI mind reading is essentially finding correlations in a unique system not any proven causation. It does not mean the pattern causes the thought any more than for example saying “Ow” causes pain in one person but saying “F-ck” in another person who likes to curse causes the same thing.

[@ArtOfTheProblem]

yes this was just a way to get the point across, definitely a simplification. but I wouldn't be surprised if we can already do this with correct AI model applied to the datastream. the patterns don't cause a thought the patterns are the thought.

[@doloresabernathy9809]

@ thats a big philosophical leap .. not unusual but i don’t agree with it. Just as correlation is not causation correlation is not identity. Thoughts are subjective experiences. How do we get from patterns in the brain to subjective experience? nobody knows. that’s why I felt Daniel Dennet’s “consciousness explained” was a non-sequitur. Some call it “nothing buttery.” a wave is nothing but water, but there’s a bug difference if your the Titanic. As of now i think it’s more accurate to say we don’t know what thoughts are or how they are generated. But we have observed correlations in some circumstances. A minority but distinguished group of neuroscientists believe that the idea that the brain generates consciousness is less supported than the idea that the brain mediates consciousness sort if like a very very complex interface between the physical world we can describe with physics and something else that is what consciousness is. The opposite may also be true but its not at all settled. In my opinion.

[@ZER0--]

AI still can't drive a car after decades of training, and millions in investment. A person of low intelligence can do it in a week. A computer can not feel, or have emotions, which is a huge flaw when it comes to AGI, (not to be confused with General AI.) It doesn't understand humour, sadness, desire, fear, etc. And ChatGTP is awful at cryptic crossword clues too.

[@ArtOfTheProblem]

have you seen the latest results on driving, it's actually remarkable

[@lord_of_love_and_thunder]

@@ArtOfTheProblem No its not. Tesla’s ‘FSD’ is so wrong on even routine driving tasks its not funny.

[@stuartdriedger]

In my experience, people don't really follow patterns. Humans experience patterns. There's a difference.

[@joshcreegan8816]

My experience has been that people interpret things by recognising patterns. I think if we didn't follow patterns in our behaviour and thoughts, we wouldn't be able to identify with each other.

[@rey82rey82]

Imagination machines

[@ArtOfTheProblem]

oooo that's a cool title

[@markburton5318]

Hey Bret, try watching Yan LeCunn on objective-driven AI. I think it helps understand what GPTs are and are not. GPTs are more like evolution than reasoning. Playing chess after watching 44 million games is not intelligence, it is evolution of chess instinct in a chess universe. When you can explain chess to an AI that has knows nothing of chess and play a novice game after a few games, that is intelligent.

[@ArtOfTheProblem]

Yes i've been following Yan. Also closely tracking experiments that do exactly this (try to teach pre-trained models new things)...however this argument feels like it's not going to hold water much longer (since the latest LLMs can indeed play novice games you teach them, try it!)

[@markburton5318]

@ yes, the introduction of some iteration through tags and RL optimisation of that process is making some difference in evaluation scores. I have tried coding, a maths set coverage problem and thermodynamics problems and I don’t get such good results. There is a lot of chatter that eval scores get better but user experience does not. It seems to depend on formulating the problem like a maths/physics exam question. It could be to do with preventing proof of copyright infringement which could be leading to suboptimal output. Most novel games are finite state machine problems on which these models are heavily trained. My view is that more science, less hacking, will be needed for intelligence. I don’t think we know what it is. There is definitely some progress here as well (eg as width tends to infinity, deep NNs approximate Bayesian estimation). But what if a mathematical or scientific breakthrough is required? We’ve been on the cusp of understanding physics since 1915 but we are no closer.

[@purezentity6582]

now, this I believe. because I had been encoded human knowledge all these decades.

[@michaelpapadopoulos6054]

Thank you for helping me see more clearly how Eliezer Yudkowsky's level of worry for this technology was right all along.

[@ArtOfTheProblem]

My goal with this video was to show both sides as clearly as possible, without taking a side. so thrilled to see it helped you evolve your thinking either way - thanks for sharing

[@Interstellar00.00]

FGAP an FGAR forever live

[@chrimony]

We are building our replacements. We will not stop. This is why Elon wants to merge with the machine.

[@RestlessBenjamin]

But we've always been "building" our replacements...new ones every single generation.

[@chrimony]

@@RestlessBenjamin Humans, not AI.

[@izzymosley1970]

Couldn't a human learn like an AI if they followed the same rules? And wouldn't it be possible for a human to learn an AI's strategies so they could beat it by using counter strategies?

[@ScientistBaffled]

Evolution has been thoroughly dismissed. The new man has no need for these archaic notions of chance.

[@kernel8803]

"I'll watch the first 5 minutes and finish the rest later....(@7 minutes) Why is this video so good!?!?"

[@ArtOfTheProblem]

!!!!! i really worked hard on this script and edit so this note means a lot

[@Madayano]

👍

[@martinross6416]

Why though? Why do we need this? Besides more money.

[@alkeryn1700]

the way llm "learn" is not even remotely comparable to how the brain works.

[@ArtOfTheProblem]

at a high level it's 'inspired by' , but mimicking the human brain exactly was never a goal except jeff hawkins, i wonder what he's up to btw

[@alkeryn1700]

@@ArtOfTheProblem inspired by yea, but i don't think you can even compare them functionally, the brain is quite a complex machine. i should check up on, him. what i'm curious about is if the transformer architecture will be used to predict / approximate groups of neurons's firing patterns and how much data capture loss you can afford before the prediction becomes unreliable over n time.

[@Tubeytime]

negative sentiment detected !

[@alkeryn1700]

@@Tubeytime haha yea, i mean even negative sentiment helps creators with the algorithm, if i truly disliked the video i'd have not reacted or commented. whilst i have disagreements with it, i do respect the work put behind it !

[@Tubeytime]

@ A disagreement on its own only makes me curious about what caused it. I know you said the brain is more complex but there's clearly more background missing.

[@mad4dam]

🏴‍☠️

[@Novastar.SaberCombat]

I loathe just about everything regarding "A.I.". Ultimately, it's a runaway *luxury* train which mankind seems to collectively enjoy, but they seem ambivalent to the looming abyss into which the train will eventually fall.

[@ArtOfTheProblem]

where do you think the train is headed?

[@Novastar.SaberCombat]

@ArtOfTheProblem I *literally* just typed it. 🙂 Did you read my comment for comprehension or did you simply glaze over the letters and scan punctuation marks? 🐲✨🐲✨🐲✨ "Before I start, I must see my end. Destination known, my mind's journey now begins. Upon my chariot, heart and soul's fate revealed. In time, all points converge; hope's strength resteeled. But to earn final peace at the universe's endless refrain, we must see all in nothingness... before we start again." 🐲✨🐲✨🐲✨ --Diamond Dragons (series)

[@ArtOfTheProblem]

@ i wana know more about the abyss

[@Novastar.SaberCombat]

@@ArtOfTheProblem It's far more psychological, mental, spiritual, and emotional than it is a physical one. That stated, the most dangerous part about the idea of any given Black Hole is that you wouldn't even know you were in one until you'd already BEEN in it for quite a while. It's a little akin to "having cancerous tissue"; at the start, no one knows. Not even the host. 🐲✨🐲✨🐲✨

[@jamesmoore4023]

Yann LeCun is pronounced as: "Yahn Luh-Koon" "Yahn": Similar to the French pronunciation of "John," with a soft "ah" sound. "Luh-Koon": The "Luh" is soft, and "Cun" is pronounced "Koon," with a slight nasal tone typical in French. Yann LeCun is a French computer scientist, so the name follows French pronunciation rules.