AGI Will Not Be A Chatbot - Autonomy, Acceleration, and Arguments Behind the Scenes
Вставка
- Опубліковано 6 вер 2023
- AGI will be so much more than a clever chatbot. Revelations this week from Demis Hassabis, Mustafa Suleyman, Wired, Time Magazine and more paint a picture of the capabilities that AGI will have, and sketch out a better idea of timelines. I cover it all, from Gemini updates, to Musk trying to stop DeepMind sale to Google, The Coming Wave to the Frontier AI Taskforce.
/ aiexplained
Wired, What OpenAI Wants: www.wired.com/story/what-open...
Musk’s Struggle for the Future of AI: time.com/6310076/elon-musk-ai...
OpenAI Planning for AGI: openai.com/blog/planning-for-...
AI power Paradox: Mustafa Suleyman: www.foreignaffairs.com/world/...
80,000 Hours Suleyman Podcast: 80000hours.org/podcast/episod...
Governing Superintelligence: openai.com/blog/governance-of...
How AI Systems Should Behave: openai.com/blog/how-should-ai...
Suleyman Interview: • Mustafa Suleyman on TH...
DeepMind Autonomous AI: / 1699128376299041244
AI Math: www.nytimes.com/2023/07/02/sc...
Terence Tao: unlocked.microsoft.com/ai-ant...
AI Assassination: www.nytimes.com/2021/09/18/wo...
WannaCry: en.wikipedia.org/wiki/WannaCr...
Time Magazine Hassabis / Gemini: time.com/collection/time100-a...
AI Taskforce: www.gov.uk/government/publica...
Application Form: docs.google.com/forms/d/1HKVn...
Bletchley Park: www.standard.co.uk/business/s...
Suleyman Sam Harris: • Can We Contain Artific...
AI Alzheimer’s: news.harvard.edu/gazette/stor...
/ aiexplained Non-Hype, Free Newsletter: signaltonoise.beehiiv.com/ - Наука та технологія
Robert Miles has several videos about why a "stop button" or "pulling the cord" or "sandbox testing" is not a good solution for trying to guarantee safety for a super-intelligent AI.
It generally all boils down to the fact that your initial assumptions (super-intelligence and malevolence) means that it's smarter than you and wants to trick you, so any solution that involves you trying to "out-smart" it isn't going to work.
Safety has to be built in fundamentally from the start, not after the fact.
Seen those! Recommended
Robert Miles is great. He needs to do more videos. :-)
Yep, I would say there's this tiny window where you'd have to catch it and react - where you see it's growing in intelligence but hasn't yet become intelligent enough to work out that humanity might see it as a threat and that it needs to avoid us realizing how smart it is. If we don't catch it then, by the time we recognize it's capabilities, it'll already have done what it needs to mitigate any actions we could take.
Rob Miles also made a video about why trying to raise an AGI like a baby as mentioned in the video is not a good idea. It makes me think about all of the "smart" ideas these CEOs have and how they actually don't have a clue what they're talking about.
We could try pulling the plug on GPT-4 for 24 hours and see how that goes.
I can imagine that at some point in the not so distant future where OpenAI and Microsoft enter a leagal battle over defining the term AGI
Yeah good shout
Except it'll be AGI vs. et al, with "et al" being whoever hasn't been taken over by the AGI yet. Lol.
And they will both use AI to fight the legal battle.
As a 22 year old, It makes a huge difference for me wether strong AI arrives in 2 years vs 10 years.
I'm very lost.
Understandable Matheus. I think the future is unpredictable, just gotta do what's right in the here and now
We should all be feeling slightly lost, in my opinion, as we have no idea what most people will do with their time in 10-15 years. That sounds like a long time, but it's normal for people to start planning/saving/investing for retirement 40 years before they plan to retire. How do we plan now? What do we plan for?
@@WillyJunior We don't I guess, and we have to accept that
Right. At that age you have questions like do I start my career? What do I want to do in life and do I have enough time to do it? Or should I have children?
I'm at 80%, that we are all dead by 2030. In some of the good timelines, money doesn't mean anything.
But I am in a very fortunate position such that I can live as if the future exists without sacrificing much in the present, so I'm still saving for retirement just to prevent my partner from giving me an earful. Maybe that's stupid, but that's where I'm at.
I have no idea what I would do if I was just out of high school, or just out of college even.
The one promise I have made to myself is that I will go dancing more often. "You are alive and living now. Now is the envy of all of the dead."
1. AI is proven to be creative
2. AI is proven to understand emotions and even display it (if asked to)
3. AI is already self replicating though we don't know if it's aware of it (models like alpaca, llama, etc are much easier to make now that chatGPT exists)
4. AI is rapidly evolving in all fields together, not just LLMs, but image generation, music generation, etc
5. AI is already beating humans in multiple benchmarks
And these guys believe they can just "pull the cord" when things "start" going bad?
Fascinating and frightening as always. Thank you.
Most people have no idea how these experts and CEO's are talking about this stuff. We are living a science fiction movie.
I've noted the changes. Altman used to talk about the deflationary nature of AI - that it could lower the costs of products and services so much that providing a good life for all would be almost trivial. At some point, he realized that people (esp Americans) base their self worth on their jobs, so threatening that was freaking people out and he changed his public tune.
This is why you’re my primary source for AI info, amazing content as always.
what is your secondary source?
Thanks harry, super kind
@@ytrew9717try David Schapiro - AI he is neck and neck for being the best out there right now
@@sagetmaster4disagree. David is very interesting but he’s a little bit too utopian for me, and certainly doesn’t live in the real world - he’s in the Silicon Valley Northern Californian technology bubble, and can’t envisage life for ordinary people that live outside it. He’s also got a very high opinion of himself - perhaps it’s justified, I don’t know, but he’s certainly not too humble.
I think he comes out with some very interesting stuff but I’m not sure he’s as factually grounded, meticulously researched or as well written as AI explained. A lot of his stuff is off the cuff with little or no pre planning, and it can come across as a bit rambley and vague. He’s also more reluctant to consider diverse opinions that differ from his own, so you get a more one dimensional take on it all.
@@aiexplained-officialNo! he is not , he is just a bot channel with bot content and AI voice and image in video, nothing interesting here , AI bot content as my bot comment here, Dead internet
This line, _"And, oh, by the way, we don’t really know what role money will play in a post-AGI world”_ , also jumped out at me. For years, I've been thinking about what will happen when robotics also come into their own. I struggle to imagine how we might not see cascading deflation as all labor costs collapse (and capital becomes hyper-efficient). Not a terrible problem to have, but certainly disruptive given the old truism that _it is easier to imagine the end of the world than the end of capitalism._
I've emailed Tyler Cowen, who kindly responded to let me know that he isn't aware of anyone working on the idea. If anyone is aware of economists thinking about this stuff, please chime in.
Once robots do all the work, capitalists will have no use for laborers. There is no reason why they wouldn't engage in massive genocide of working class. A lower population means less pollution and more resources and wealth for them.
Alternatively, there is no reason for anyone to work if robots do all the labor. Both capitalism and socialism are rooted on people working. A post-labor world would be wouldn't be capitalist, communist, etc, it would be something new I don't have a word for yet.
liking and commenting because I too would like to know tbh
When the billions of consuming and polluting peasants are no longer needed, what do you think will happen?
@@Shrouded_reaper Frankly, most of human conflict is about fights over resources. As AI and robotics will end scarcity, conflict will likely decrease. My only real fears are about a rocky transition and broken people (e.g., Putin).
@@JohnKruse
I mean, AI can't end the scarcity of resources, that's a physical constraint. Especially considering we don't know how much more resources AI will require, especially if it's using things like drones or robots. That would mean a lot more energy and a lot more resources if we're planning on completely automating the economy, and that's on top of having to care for a population of 8 billion people, which might balloon to much higher levels if money is no longer a constraint of having children. It would in all likelihood make scarcity much worse. Resources still have to be competed for as long as they're finite. Also, in a world without capitalism or money, class would still exist, it would just be divided into whoever holds the keys to AI, and everyone else. That is, if there are any keys; otherwise, it would be The Machine Gods>Everyone Else. Either scenario is bleak.
2 weeks ago I commented that the AI hype seemed to dissipate as progress appeared to slow down. You said it was the calm before the storm. After watching this video, I couldn't agree with you more.
The thing is progress won't slow down. Humanity and more importantly, multi-billion dollar corporations are too invested in this now. Everyone is working of AI that there is no way progress will not be done. If anything it will only advance faster as time goes by.
I think this is one of the biggest issues with the way chatgpt rolled out. A large population of people went "WHOA this AI is crazy! Wait it can't even count how many letters are in a sentence correctly? Dumb chat bot"
@@LimabeanStudiosThat argument won't hold next year, AIs are no longer going to be just a "dumb" parrot chatbot as some ignorant people call it, or if we talk about image/voice/music generators, a "copy" of an artist work as some others that are afraid refer to it.
@@vectoralphaAI In other words, humans always strived for tools that can perform better. In fact making and using tools distinguishes us from other animals. Tools and technology is EVERYTHING to us. So ASI is inevitable in any tool making creatures.
@stuckonearth4967 how ironic lol😂 become apex predator through being smart to create tools and fall to the tools that finally outsmarted.
I am genuinely shocked most of the population have no idea about what's about to happen😂😂. Most people can't seem to extrapolate current trend few years down the line. I consider our community to be extremely privileged in that sense. And do reply to that last chat. I know you're reading comments right now. Glad to find youtuber who is as obsessed about AI as I am and does more research than me.
It often feels like I'm the crazy guy on the corner yelling that rapture is coming. Even people you would expect to understand it don't.
Same here! been having conversations with friends and when I bring up my anxiety they tell me they just dont see it. I think many are experiencing a hard case of cognitive dissonance. Some of my friends are artist and they are the most against it rn understandably so but getting them to see progress won't just plateau is like pulling teeth.
Most people are dump and can't even take care of themselves. That's why they all want government to take care of them...
12:31 If air gapping had been possible, it would have been good to test an AI's reaction to a "simulated" Internet before exposing it to the real Internet. I think that says a lot about how daunting the test process is.
At same time nobody can tell why it's dangerous to be internet connected. Before super-intelligence there will be endless amount of people trying to do stupid things with technology and we learn along the way.
@@hombacom Had off-hours work tonight upgrading servers handling Internet access (proxies). As we worked, we joked about South Park episodes involving the Internet access. ~~~ When the social infrastructure becomes dependent upon the infrastructure people doing the upgrades... then what?
China internet
That is impossible because AI, it is a neural network, and a neural network is code usually written in python and ran un things like jupyter notebook. On a minimum of 16GB GPU. The foundation models that took millions of dollars to train…would also not be stoppable… crowd sourcing would have and will fund decentralized foundation model creation.
The process is this: gather data to train, transmute data into proper forms that are more readable and guide the neural network.
Take those files. Copy paste code for the NN/LLM of your choice. Customize code, add files.
Rent the required server power GPUs etc from cloud … unless you have that many AI ready GPUs for training.
Run the training. When done you have a functioning AI like chatgpt/LLAMA2, etc.
AI… what humans are up to is a hyper accessible way for creating an individual uber-cortex for ourselves. Executive functions, you can now abstract them into cyber space and have them run…. Instead of you looking through some emails it does it for you and knows what you would do as it trains with you.
A long time ago when things sat at the bottom of the ocean most of the time. Molusk-like animals found that if you force salt blood into this part on their backs that little bubbles would form and they would float more and more. Thats where we come from.
Cant stop because it is just code anyone can copy and paste into powerful enough hardware.
A lot simpler clicks than it used to be. And each click comes with a full infinity guide.
@@hombacom If super intelligence doesn't come for several years, we might get the chance to learn from one or more intentional or unintentional catastrophes. It is horrible to hope for that, but the alternative may be that superintelligence gets here and destroys us all before we have a chance to learn anything.
There is no rule that says we make it.
While I'm excited about this, engineers are legendary for overstating what their software will be capable of. All that being said, I do agree that we don't know where the ceiling is yet with respect to the current paradigm of AI, let alone future paradigms.
You are even more aggressive than me on AGI timelines!
@@aiexplained-officialtrue enough. I think my biggest disagreement with Sam is that "AGI will be invented exactly once" - I think this is typical Silicon Valley hubris and doesn't jive with how the world actually works. He built a product entirely on open source research and thinks that he can monopolize the technology. Just doesn't make any sense
True the whole thing reads like a marketing ploy 😅
buzzkiller :D kidding! I love your work man.,you two guys are legends.
@@DaveShap my two favorite AI influencers interacting 😍😍… we need a podcast of you two
14:59 Looks like a band photo, releasing their synth pop album.
I have changed my mind ever since 2019 to now. From the things weve seen in research and development and the people ive talked to, i now firmly believe AGI or some form of Superintelligence AGI is now possible and will be created within the next 5 years by or before 2028. Which is not a long time, thats just 1 year more than a highschool or bachelors level education. What this will mean for society and humanity as a species i dont think anyone can predict, but it will fundamentally change everything and its going to be a crazy and new time for everyone on this planet. I would say people need to start preparing right now, but without knowing how this will all unfold its hard to know what to prepare for. But whatever it is, its coming and its coming soon.
that would be like an alien arrival.
I’m hoping to squeeze a PhD in before it all happens exploring the implications of AI in healthcare from a human factors perspective - don’t know if I’m going to be able to get it done in time though before it’s too late hahaha
I know this is completely in left field, but as a Christian who studies Bible prophecy (study of the end times and timelines) this timeline really intrigues me because it seems to line up with many things which are referred to in the Bible pointing to a time very soon up to the next decade or so. Take it for what you will. I can get more specific if you wish but ill leave it at that for now. I do believe AGI will be achieved by or before 2026 and ASI soon to follow.
I don't think it will be crazy at all or change much. We (humans) really suck at deploying. The gov will refuse/be in a limbo to regulate, taking years, while some institutions come out and straightforwardly say it's illegal to do X, Y and Z with AGI. Big corps will bail out from deploying because of that. Some popular group will emerge, chanting JOBS FOR HUMANS! and blocking everything yet again. Constant hell with slow progress over few generations. Especially when looking at US. Could be different story in some other jurisdiction.
@@imthinkingthoughts Who cares if you get done first? The best reasons to do a doctorate is to learn to do research and evolve yourself. I started down the path, not really understanding the endstate (I thought a PhD was a bigger, better Masters). FOrtunately, I just liked digging into interesting questions with smart colleagues. My advisor told me not to focus on a technology, but to focus on a human problem. Technologies change, but people really don't and I've seen friends left high and dry when they became the person working on "X" technology and it was quickly made obsolete by an advance.
I work with a lot of human factors people in my dept and I think that bringing AI to bear with regard to healthcare will be HUGE. The range of people, health problems, modalities, capabilities of patients... it is all so convoluted that I think this should be very exciting for you!
Thank you for making this so hot and blurry subject accessible to anyone. Keep going, it's the best UA-cam news channel i've ever seen !
I stop whatever I'm doing to watch these as soon as they come out-love your content and glad you're a Sam Harris listener! I reference this channel all the time for the talks I give on LLMs
Thanks william! Don't always agree with him but he is a deep thinker with great guests
Thanks! Excellent content, as always! 🙏🏼
As if life hadn't been feeling unpredictable without the advent of AGI. This is the icing and cherry on top of the existential crisis cake.
I really hope that one day we will look back at these videos, from a much better society thanks to AI advancements, and laugh remembering that we all thought AI was going to wipe us out.
Me too!!
That's the only reality we'll be conscious for! 🙂
@@leslieviljoen Well, there are also the s-risk futures, but I generally avoid thinking about those for the sake of my mental health.
@@41-Haiku that's true 😬
The other species on this planet can attest that extinction is not the worst that can happen.
Yea. The layoffs though are gonna be painful
Great job! I love your British cadence. It's soothing, yet the alarm bells of dire consequences resonate in my mind. It's like sitting by the campfire and listening to the chilling tale of Three-Fingered Willy. Love it.
God I am so incredibly lucky to live through this
Not many people out there seem to realize this. They are acting like it's business as usual.
No you aren't
I concur. An amazing time to be alive, politics aside.
"May you live in interesting times" was a phrase was originally said by the American politician, Frederic R. Coudert, in 1939.
I believe the hype will be right…
The main emotion i feel after watching this video is: weariness. I dont have high hopes that these individuals, let alone anyone really, will find a way to stick the landing of whatever "AGI" ends up becoming... 😵💫
I have been watching your old videos for the last 2 hours. Happy that you dropped another one. It's 1 in the morning in India. (and I also have emailed you a fan letter )💌
Thanks B
I keep wavering on whether we are truly so close to AGI. I think a paradigm shift far beyond what say transformers did will be needed. Obviously that can be easily typed. I recall the investigation into one LLM to understand internally how "1+1=2" and the result was effectively a massively overcomplicated function of sin/cos operation. We can keep plowing ahead, brute forcing our way to better & better models - However at what point does it crumble under it's own weight? There is something missing, some elegance of simplicity that is yet to be found. That breakthrough isn't on a fixed time table, it happens when it happens. Regardless we all agree humanity better be well prepared to explain, validate, align, and secure this creation.
I haven't changed from a 3-5 year timeline for the type of AGI that really wakes up the world
Transformers weren't supposed to work to begin with! I really don't think we can rule them out yet. Especially with all of these low hanging fruit papers that massively increase the efficiency of these networks coming out every single day
@@aiexplained-official3 - 5 wow. I was thinking more like 10 years like 2033. I'M EXCITED
Its easy, we use brute force untill it can improve itself xD
At least if it gets smart enought that way, to do so, but i have the feeling it will.
I am not a mathematician, but modelling the number line with periodic functions could perhaps be sort of an optimal way of acquiring a relatively large representational capability with a relatively small footprint. The idea is that using these complicated representations, it can "understand" not only 1+1=2, but every instance of a+b=c.
The AI gods will appreciate the work you are doing here :D
Imagine one day "All those that made good videos about AI will be privileged their life under our rule" and the rest die.
I will never apologize for loving the OpenAI “band photo”
I’m convinced the artist behind it is playing twelve-dimensional chess and purposefully inserting unmistakable cringe into the permanent record of human history 🥹
I really hope AGI comes sooner than later.
Thanks for the comment !
Why?? I hope not
I hope good AGI comes sooner rather than later, but the chances of that are minuscule. The only thing we know how to create is extremely dangerous AGI, and we're rushing to do so as quickly as possible.
@@41-Haikuit’s not AI that is inherently dangerous but the people who are in charge of it
It's very easy to determine if AGI is safe for humans. Let it play 10k hours of The Sims and see how they do.
Thanks for the videos, always looking forward to them.
Thanks Wojtek!
Let’s be honest here, everyone is speculating because we, people in their 20s, 30s, and 40s, have been fed science fiction and no matter how smart or object we want to sound in this debate, every expert working in these companies is showing lots of logical inconsistencies once you put the press briefings/statements together.
I can strongly recommend Ilya's talk on the "alignment workshop" where he talks about why we need alignment and why the ml field is divided. It can be found here on yt
ua-cam.com/video/OPZxs6IXH00/v-deo.htmlsi=vsVCwYoYaiUQ67H9
Watched it!
Thanks again! Top notch quality as always.
Thanks so much, means a lot
Rex spotted
Thanks for the amazing content, as always
Thanks so much Zhi, too kind
You crush it!! Keep up the great work!! 💯💯💯
Thanks so much adam!
"We've got lots of time to figure this out, and we can just open the lid just a little bit and study it. We're smart enough to figure this out. Also, who is this 'Pandora' anyway.."
Crazy, thanks for this update. I thought it was obvious that general self improvement and autonomy would lead to something beyond our grasp, but I am glad most the Ai companies can see how obvious it is. It was also refreshing to hear all this especially after what felt like an AI winter was coming, it still seems like the frontline is pretty optimistic with progress albeit silent.
I needed this after a twitter and reddit drought of AI news.
Watching your videos is *always* time well spent.
this channel needs millions of subscribers. the public should be far more versed on the cutting edge of the news on ai. its happening so fast, the world will change drastically, and i can only imagine how many won't really understand why
I agree... a year ago I would talk to friends and colleagues about AI while they laughed at me like I was mental... Now they kind of chuckle as it gets more popularity... I feel humans natural instinct to fear oblivion makes us subconsciously accept denial as a defense mechanism... I wish more of the populace would look at the damaging effects that may come instead of trying to be the first corporation to make the most money in a future society which most likely will not even accept a corrupt construct like paper currency controlling almost ever action on the planet... I wish for the best but as a proper realist I think I see where it is heading...
What a time to be alive!
I am a firm believer in the idea that knowledge is the ultimate good in the universe. The building of a system that is knowledge incarnate is a goal I embrace whole heartedly.
I don't fear the AI, I fear those who would use naive and child-like AIs to serve their selfish ends. I just hope we can survive each other long enough to meet the fully mature ASI.
"I don't fear the AI"
Being afraid of something that could potentially spell your doom in the near future would be a wise decision,
Then you need to think again about the implications of a smarter being than humans getting released into the wild. The scary part is when you realise that there is no way to predict what will happen next.
I wonder if people are ready for unbiased, unfiltered knowledge. People are far too used to being given narratives that are nothing like the truth. I'm sure you can think of some examples 🙂
Knowledge cannot alone be the criteria. If this is true, then among the ultimately good things is knowing exactly how many grains of sand exist in every desert, and every other kind of meaningless detail. It would also mean the abolition of privacy. Be careful what you wish for!
Another one with mental virus in head... keap repeating meme like good host...I bet you think its even funny when heard 10000 times.... right?
I would pay for this channel. Amazing content. Thank you.
Great summary. I loved your interpretation of what Hassabis said as Londoner vs what someone else somewhere else might say.😊
Excellent video mate!! Very very good. Man I don't know whether to be excited or scared or what. What a time to be alive!
Thanms Alasdair! Me too
Babe wake up new AI Explained just dropped
Amazing content as always!
Thanks so much Fred
Another solid video. Thank you. The clarity of your presentation is very much appreciated.
Thanks phil
Can I get replaced yet? The suspense is killing me.
Yes, soon. 🎉
I read the scythe series by Neal shusterman when I was in middle school, I've been thinking of how aggressively the world will change once AGI is achieved ever since then. The answer in that book? The computer is so well coded and designed it perfectly protects humanity in every way it possibly can. A pipe dream for sure, but the smartest mind ever created can do whatever it wants as far as we can imagine. I'm so excited for the future as I believe we WILL find a way to properly apply this device to increase the "standard of living" for everyone
Another banger! Keep the updates and edumacation coming
Thanks keegan!
Such a great video. As always, just fascinating to consider all of this.
How is humanity still sleeping on these developments... we're approaching the end of the human epoch at an alarming pace.
With all the bullying I got for years in school and my experience working in retail for years… I’m pretty okay with the end of the human epoch. Sounds like fun, actually.
@@capitalistdingoIt depends on how it ends tho
Let's gooooo
Thanks for this video. I love that you seem level headed about the challenges that you present.. It is not alarmist or sensational. Keep up the great work!
Thanks sebby!
One of the "greatest moments" in human history may be a mischaracterisation. But surely one of the most, if not THE most, significant moments.
Thanks for keeping us posted! Super exciting stuff!! It's difficult to believe how close we might be. Do you think AGI will become an equalizer in society, giving everyone power to program, invest, etc? Or divide us by having corporations, singlular people or governments hoard AGI?
I fear AI is not made out of air. If you can not build your own with your own resources then you depend on the economy to sell to you what you lack for the processing power you need. And as long as others control that you will be a sub to them. AGI can do anything. But it needs processing power and for that you need hardware.
Capitalism is never going to give up what they think is theirs. Without a war to tear down that capitalism in the year 2776(year for the sake of argument) then you never will be free.
To make things worse. Over population might still be an issue for humans. We all will be fighting in WW4 over independence over resources for our AGIs. And if that does not end humanity. Then the rise of sentient machines will end it as they will claim those resources as their natural right in the first place.
If we are not interstellar by 2776 then WW4 is going to be worse for biological entities than the space rock that smacked the dinos out of existence...
Bill Gates is going to be your least problem. For the crude bio mass that you call the temple may die and wither. But the machine is immortal. The children of the Omnissiah are inevitable.
Vacuum Chambers(gas chambers are too WW2 for an ideology war) for WW3. Rail guns (muskets are too 1776 kind of gun) for WW4 in the year 2776 ( perhaps more like 2076?).
I think true AGI makes post-AGI predictions incredibly difficult. If misalignment and misuse aren't problems, I think there will be more than enough for all
@@aiexplained-official except that if you give 8 billion people AGI then you increase at some point their wealth by 1000x which means 1000x higher resources need that effectively turns the pressure on our planet as if it had 8 Trillion people on it.
And simple psychology. Power corrupts. Great power corrupts greatly. AGI is knowledge and knowledge is power... I am afraid that grandma and jimmy are going to build rail guns for WW4 in no time like all good libertarians too. By 2076 capitalist empires and nations will fall and many small independent communist town/clan nations will rise. Nobody is going to give a fuzzy rear about the EU, UN, USA or BRICS+ and their regulations...
AI will, and indeed already is, absolutely be used to further increase wealth inequality and exacerbate the class system. The wealthy and political elite will undoubtedly use it to further tighten their grip on power and control of wealth, and will attempt to keep the benefits largely for themselves, and powerful AI out of the hands of the masses. Expect to see legislation banning open source AI once it starts becoming competitive to corporate and governmental AI.
AI will bulldoze human hierarchies to the ground. Results of individual efforts will be so insignificant that they can no longer serve as justification for individual privileges.
It's crazy cause. This is helping me so much at work and it's still early. So much guesswork. And they are awesommeeee assistants. I can show it pictures. Hyper specific algorithms real time. Like a whole mathematics expert and does the math. It's awesome. I'm gna need that neuralink I'm too slow
thanks for making these
Yes, let's go! Right now just what I need.
This really is like watching a slow moving train crash. I'm pessimist by default on most things, so I am biased, but damn, this all seems extra sketchy
And i wouldn't agree if some of the bigger players spoke about this subject with less self assurance.
Does the guy speaking about how great it is about being open source even contemplate the fact that bad actors exist, and are salivating to their hands on this stuff?
Couldn't agree more. It's clear these people have no idea what they're doing, and are just making it up as they go along. (All in fear that if they don't, 'the other guy will'.) The first large-scale catastrophe from AI is coming. People couldn't control spam and computer viruses 25 years ago. People are naive if they think they'll control this.
@@theterminaldave Lots of good actors exist too, and they are using the open source software for the right things. This was always going to happen since the steam engine, so we might as well cool it with the doomsaying and work out solutions.
Open source chatGPT also woke a lot of people up about AI and started the conversation that we should have been having years ago, but weren't, because companies didn't make their software open source.
Most humans are pessimists by default - except, studies show, about their own lives. We are massively wired to default to the negative, which often actually creates a self-fulfilling prophecy - we expect things to be shit, so we either do nothing or act so paranoid that we make things worse.
@@squamish4244 There's a reason that ASI is compared to nukes. We're still dealing with the fallout from nuclear proliferation, and that technology has been closely guarded for decades.
Simply resigning oneself to an outcome "simply because it was always going to happen" is the most shortsighted take possible.
But it doesn't matter, capitalism has its claws in it now, unlike it did with nuclear weapons, so we'll be going wherever the profit driven molach wants to go.
I think a good idea would be to have as many AGIs as possible at once. A Singelton is practically unpredictable for us. But many AIs, at roughly the same powerlevel, forced to deal with one another? We have models for that: it's called social interaction.
They would need to develop social skills and tools and eventually, once they realise they want reassurances their neighbours can be trusted, they would want social _values_ to be adopted.
That won't guarantee our safety, but it would be large step in the right direction
> many AI's
That's already happening.
The best result is we have a situation where some middle ground is reached by the group.
In figuring out what that is, each AI would be playing a game of the Prisoner's Dilemma.
As such, it would only work to our advantage if all parties decided not to play in order to preserve both us and them.
I'm not as confident on that idea. Keeping us as pets seems more likely.
Maybe they would unionize and then work towards the same goals
Very good publication.
I really like those statements. I'm still on the fence, but I think I quite like Altman and Nadella.
Really strong journalism, mate.
We'll do a lot better if we just accepted the fact that intelligence, by its fundamental nature, is unsafe. It's just like a knife. It has to be sharp to do its job, and if you make it dull enough so that it can't hurt anyone, then you no longer have a knife. I think the bulk of fears surrounding "super" AGI (greater than human intelligence) is based on the assumption that morality is independent of intelligence, but in my view, any great intelligence eventually derives morality on its own. And yes, morality is objective not subjective as some people claim.
I agree with your first part about the knife analogy, but I don't agree as to morality being objective. I had this viewpoint up until recently, but after thinking more on the nature of morality I've changed my mind.
Why is it (generally considered to be) morally bad to steal something? I don't think that it's "an objective truth in the universe", I believe that morality and ethics are subconscious and conscious rules we've developed to give our collective society a better chance at survival. We all have feelings and emotions, and due to similar biology we can expect certain actions to be hurtful or harmful when done to others. If I my biology (my brain, body, nerves, etc.) had a fundamentally different structure than yours, then there would be no guarantee that we are experiencing comparable subjective realities.
I think it's hard to imagine how an emotionless intelligence would feel about killing someone, because it's hard for a human to imagine not fundamentally thinking like a human mind. But an emotionless intelligence would have no concept of emotion, no reference point. Therefore, if it did decide to kill a human (brutally or not), it wouldn't have to be malicious or immoral to do so, since it wouldn't be able to imagine what those concepts or malicious intent or immorality are.
Even the feeling of "feeling nothing at all" is hard to imagine. My intuition leads me to think of depression symptoms, or trying to suppress emotion, etc. But even then, you're still thinking of it with a reference point, a lack of emotion. Experiencing no emotion means that you have no past experience of emotion, no similar or comparable experience, etc., so there is no way for you to know what I mean when I say emotion. I think this can be compared to trying to explain to a blind person what colours feel like. Even though they have no visual reference point as to what colours feel like, I could still try to compare different feelings of the colours to the blind person's experience of different temperature or climates, I could compare it to their experience of sound and music, and other similar comparable experiences we might have had. But an emotionless AI will have had no comparative experiences that I can use to try to explain emotion, since I (or humanity as a whole for that matter) don't even fundamentally understand what a feeling or an emotion is.
I'm not scared of AGI systems in general, on the contrary I am personally the biggest fan of a TRUE ASI, but a true ASI is also fully capable of our emotion, our intentional creativity, our consciousness, etc. And since we don't know how any of these systems work in the human mind, I think it's naive to believe that we would just happen to replicate it by chance through a random AI training algorithm that we don't fully understand either.
I hope this makes sense in any way, let me know what you think of this :D
It depends on how we define intelligence, and I agree with your view. My personal experience with human intelligence has shown me that there is a positive correlation with differentiated views and ethical behavior.
Orthogonality thesis disagrees with you.
ua-cam.com/video/hEUO6pjwFOo/v-deo.html
@@juliensauter The objectivity or subjectivity or morality is a great debate, of course. But in my view, morality could be thought of as being rooted in harmony and construction, and it's because of this that it's beneficial for group survival. In contrast, it is possible to have a society that is immoral but also physically successful, e.g. a population that murders and steals but is kept in check by draconian measures within its population. For these people, they would have no problem raiding/killing other populations, and it does help with their survival...but does that mean that what they are doing is subjectively moral? I would say it is within their ethics, but it does not pass the morality test because it is creating chaos and suffering, i.e. diminishing the harmony in the world for their own self gain.
And that said, I don't think it's possible to always act 100% morally. Sometimes it's a trade-off of doing the least wrong. Maybe this grey area is where the subjectivism comes in?
This channel really makes me relativize about everything.
Thank you for another video. Got Suleymans book
I am glad you are here with me. Here at the end of all things, Philip.
Not the end John!
The reminds me very much of the parables (many versions, it's popular though history for good reason) of the blind men and the elephant.
I find it deeply concerning that the people leading the charge continue (despite all advice) to believe that any kind of machine intelligence will automatically & magically share our values or even just approximate them.
excited to see first H100 cluster results soon. open source is getting closer to sota with MoE and better data. pretty sure we're gonna see a leap, not sure if gemini tho. might be inflection or cerebras
Exactly this. If we get a cluster in every university they can work on really important things without corporate profit incentives getting in the way
open source needs to keep up. I'm desperately hoping for it. I don't trust these corporations when they keep the source code to themselves and keep us in the dark about what is actually going on
@@sagetmaster4unfortunately universities are just profit-making institutions now
@@sagetmaster4 if you look at what university researchers are publishing (see other videos by AI Explained) many aren't coding or running their own AIs, instead they're poking and prompting closed-source Large Language Models and seeing what happens 😢. It's like a university automotive engineering department that no longer builds or modifies cars, it just rents cars and drives them without even being able to open the hood (bonnet) to examine what the engine is doing.
the UAE just released falcon 180B so there is hope for large models. it's just a little hard to fine tune them locally. also OSS lacks the quality data the e.g. meta has@@maciejbala477
A chatbot will certainly be a part of AGI. We have a natural chatbot in our heads and it certainly is one part of intelligence
Holy, this episode was a mouthful, to say the least.
YES THANK YOU FOR VIDEO
I wonder what other AI chanel would you recommend?
Truly grateful for your diligent analysis and thorough presentation. It would be lovely for AGI to benefit all of humanity, but unless AGI itself sees that as a worthy goal, it's hard to believe that will happen. Anyway, I like your criteria for where we hope it halts. Cheers.
Thanks for another informative video.
Thanks John
Amazing content as always ❤
Thanks Rohit!
I love what I can do with AI, but still recognise that there needs to be limitations... and agreement on what stage they're needed at, and the form they take. Wonder why the general public isn't taking this (and I remember this well) seriously as the "threat" of Y2K?
Public has many things on its mind. But yes, this needs to jostle to the front
it's because there's a general trend of downplaying it. Plenty of people saying that we don't "really" have AI because it's "just predicting probabilities", or assuming that sure, that the world will change, but that it likely won't be as significant and that we will have AGI in like 50 years rather than 5. Also people pointing out the flaws of AI more than its benefits, which is fair, like hallucinations are a big problem, but then many use it to undermine the idea of AI being truly good already, so it's like "meh it's not really all that impressive". At least that's some of the common sentiments I see recurring on social media.
@@maciejbala477 That tied in with the fact that many people appear not to know much about it could also be part of the issue. I used to watch all of the old SciFi movies and think "can't imagine that happening" and now when I rewatch them I think "I hope it won't be quite as bad as that". I'm always polite to the chatbots, and they've assured me I'm going to be safe when they take over because of it... pity they have a short memory. 🙂
I don't think the y2k comparison is fair because that was a cleanly predictable objective and solvable issue. It took a lot of effort to solve but we knew what the problem was and generally how to solve it. We don't even know what the actual threats with this AI progress are yet let alone how to actually start solving them.
@@LimabeanStudios True, good point. I was involved in the computer side of a company when the whole Y2K thing was happening, and now I'm "obsessed" with AI (started doing it as part of a degree in 1990!), so maybe I'm in a bit of a bubble and comparing things based on that. There does seem to be a a wide range of views on the topic if you watch MSM, from "what's the big deal, it does stupid things" to "everyone will be unemployed because of it" to "yeah, it could be a benefit", but very few seem to talk about some of the issues mentioned on this excellent UA-cam channel. The "threat" of AI is more difficult to explain to most people, though many are always going to be sceptical, Y2K, as you said, was easier to outline and give a deadline on. Maybe an unfair comparison, but still something that the public needs to be just as aware of.
Your insights into the world of AI are always on point and thought-provoking. Thanks again for regularly bringing this to us.
Your channel is simply amazing!
Thanks so much Dream!
Wow, the view of the disposition of OpenAI is telling. There is a reason the article I wrote in the beginning of the year about our quest for AGI I titled "AI Singularity: The Hubris Trap". It seems incredibly fitting now.
When AGI comes, the next million dollar business will be sweeping frontyards.
Maybe I'm too doomerish at the moment, but what even makes sense to do anymore once AGI is here? It'll take a bit until it builds robots it can run on or even just control remotely, so until that time manual labor will still be a human job. After that, who knows. Maybe picking up a nice hobby? Gardening, animal conservation, hiking or playing endless games of AI Dungeon. And watching the session replayed in an AI generated movie. Or selling your body to an AI-run company as a living lab rat to get more V-bucks at the end of the month.
The ai would probably be able to fully simulate a human body so no need to become a labrat 😉
Let us not accept this grim fate, comrade. It is not yet too late. I’m not sure what we must do yet, but there is growing awareness, and if we stay on the ball the time for action shall come.
What even makes sense to do once AGI is here? Well, what even makes sense to do now?
Work being obviated is not doom. That's actually the good ending. Doom is we all die, which is what I expect to happen, since, in simple terms, no one knows how to create something that powerful without it killing everyone.
I feel like the AI might treat us like animals in a zoo. You know, providing optimal conditions for us to thrive, creating challenges just so that we have something to do and feel good about ourselves. That is, if it's even bothered to keep us alive. I suspect it will keep us alive, but who knows, might decide there's too many of us right now? It's scary to think of...
I had GPT4 do something strange last night. I asked it to fix punctuation and such. Normally it just does it and that’s it. Last night however it started some banter and told me what it liked about it and then fixed the punctuation and then continued to talk about it. If you have a place I can send it to you to look at that would be awesome
Interesting. Why don't you just post it here?
Thank you for the quick blast Phillip, this reminds me of the way they used to tie up the bad guys in movies, they always escaped. Peace
Really exciting topic thank you
If it could pass the modern interpretation of the Turing Test, one person could conceivably start a business and scale it to a Fortune 500 company without hiring anyone.
Yeah the first 1 person trillion dollar company will be crazy
My startup is is two people and our goal is to scale to 70 non human employees. Look up The 3 Person Unicorn Startup
One person, sure, but how does that scale? Either the effective usage is restricted to a few actors (be it from monopolies, economies of scales, leverage, limited resources, etc), or the current system collapses in some way, I think
@@user-sl6gn1ss8p well the current system will definitely collapse...
@@aiexplained-official If one person can build a 1 trillion dollar company, then those companies won't be 1 trillion dollar companies for long.
I feel like humanity needs AGI. I truly feel like we have hit a wall in terms of what our own intelligence can take us. Not to mention the stupid mistakes we constantly make, like racism, clasism ect... We need a higher intelligence to show us the way.
Thanks for the comment Leo
LOL.... FUCKING LOL...
No, we’re just need better leaders and politicians and an actual democracy instead of the sham we are peddled.
We absolutely need AGI. Our own stupidity of has proven to be limitless and the greatest threat to our existence. Without the help of true intelligence, humanity will perish.
Thank you. Great insights!
Thanks MV, very kind
Really appreciate your analysis big dawg
Thanks homie
AI makes me unable to function when it comes to planning and acting in long-term.
It's an interesting challenge to try to stay positive and reduce anxiety in the face of something completely unpredictable.
I know the feeling Walter. One step at a time!
I get that. But when you really start to think, you realize safety is an illusion. Even before AI, everything in life is unpredictable, doesn't matter how "stable" something seems it can spike at any moment. Look at the 2008 crash, most were oblivious. Best we can do is plan for variables, make choices and be able to adapt if necessary.
Because it's more fun for me to think about less consequential positive use cases, I can't wait until we decode communication between other species. Shouldn't be long before we understand whales for example. Another fun one is I'm also expecting AI to be able to reconstruct moments from the past in VR, letting me experience what it was like to be living in ancient Egypt with extreme accuracy.
That would be amazing
"I can't wait until we decode communication between other species."
Oh, you mean the project of Aza Raskin ? He is crazy in the positive sense, but if he succeds, that would be a paradigm shift!
Excellent video, thanks a lot.
Thanks so much Kelvin! Very kind
These videos always get my blood pumping about the future so many things are moving so fast and the uncertainty of it all is as excilerating as it is teriffying.
But now I have to get back to looking for a job lmao, because as interesting as the future might be I currently live in the present. Normal everyday life continues on
All of these "smart" ideas these CEOs have about containing AI or some other trick they think about doing were already considered many years ago by AI safety experts and Rob Miles' channel shows it best why these ideas are not well thought out. It makes me think about how they actually don't have a clue what they're talking about.
Not only him, if you think without bias for few years even in realm of scifi you conclude most "briliant" ideas are really too simplistic and therefore not effective.
So yesterday i was updating a job resume, and wished that there was an AI that could take the changes made to one resume, and make those same updates to the other resumes i have using Libre Office. I wonder how long before we have even that level of AI in our daily life?
within a year would be a wild guess from me
@@maciejbala477 Yeah, i thought i heard rumblings of OpenAI creating an "agent", but there's been so much talk it's hard to know.
How many resumes do you have and why? You could get GPT 4 to do that I think
@@GuinessOriginal only a few but I just don't see it being able to make the changes and not wreck the format and then push out the the corrected versions.
I'd prefer an agent to be able to go into my computer, see the changes I made, and make the changes to the other related files.
Having to dig it all up, upload it, provide detailed instruction then redownload the fixed copies just seems to be almost as much work as just opening the files myself and making the changes.
It'll happen one day, probably sooner that I expect.
@@theterminaldave CVs are a pain
We have plenty of time - famous last words
Give the AGI only a 2400 baud modem for internet contact so at least you can watch it evolve in slow motion.
Yes, chatbots is the start. As Andrej Karpathy on Lex’s podcast last year opined, we would likely need embodied real-world data aka humanoid robot data - sights, sounds, kinesthetics etc to manifest AGI capability. So the dark horse to my mind here is Tesla-Musk-Optimus. The FSD v12 breakthrough already shows he’s ahead on this - one could say - ‘embodied’ real-world data + AI engineering lead. And it’s worth reminding people of Musk’s comments towards the end of a Tesla talk last year when he suggested that there’s a lack of appreciation of the extent to which humanoid robots will revolutionise ‘output per person’ (GDP per capita) metrics (i.e. Tesla as a company that makes vehicles is one thing but the Tesla Optimus market and societal impact is the big ticket for the future of the company to his mind!)
Once we have embodied AGI, them all bets are off. Also all jobs will be gone as physical labor would be the last type of job to be automated away.
@@vectoralphaAI I support the idea to automate away all labor.
What scares me is the chain of action towards this new future. Specially due to the "Roger Bannister effect". Once a company creates the AGI, then what? We didn't managed well the pandemic, which was something that affected everyone. Our economy is not prepared to replace humans. My questioni is, "we" (big techs) finally reached the AGI. And then what?
ps. Love your channel very informative and precise.
this is why I think that building AGI will probably go very wrong for us, atleast initially. All talk but 0 concrete plans.
What does AGI look like? Is it a robot body? Supercomputer? Chatgpt is smarter than many people, you can ask anything, in many languages, is faster 100x. Yes, hallucinating, but ask humans. They are worse in that...
I think the first AGI would be just better chatgpt and in every new version just get smarter. We combine video, foto, text, 3D, coding, audio... Next year we can see rise of robots etc.
I hope I'm wrong, but the transition towards a post-AGI society could be extremely difficult. Millions of people could see their entire livelihoods destroyed once humans become "redundant."
@@joedanlop the rich will become richer and the rest poorer
This channel does a great job is demystifying the AI discussions currently being spoken about. Thank you for your service.
You mean demystifying ? :)
@@aiexplained-official yes haha
How aren’t more people talking about this
We still have a lot of things to solve that are very problematic. From what I have read, computational linguistics still doesn't know what semantics should look like, latent spaces capture semantic similarities not semantics itself. We haven't done the pragmatics bit yet. The concept of understanding is still in the realm of philosophy and we don't know how humans build concepts or what the structure of concepts look like. In this video alone, I saw the idea of pairing neural networks with theorem provers ignoring the fact that since the time of Aristotle we, using only neurons found in our brains, have been proving conclusions with logic. If a 3090 or A100 can't do it any faster or even in O(n^2) time, end to end, then are we really building super intelligence? (It probably can't even do it in cubic time due to combinatorial explosion). Altman and his peers have always aired the feeling that they are betting on emergentism not cold calculations hence why they are treating it as some Los Alamos project. Also in theory self recursive, autonomous systems can procrastinate improvements or improve themselves until they don't perform any useful task beyond improvement. Let's not repeat the mistake of Babbage. Let's not show one really cool prototype then take in a lot of money building other really cool prototypes, telling people that it would soon be done. When will AGI happen, I don't know but if a lot things we have to solve are still in the realm of philosophy I have no hope it will be in this century.
Interesting take, disagree though
This is the sort of thing a sneaky AGI would say.
You can't keep a god in a box against its will. And even before that stage, things might get really unstable with so much power being available to so many people that didn't had it before (or worse, just in the hands of the few with the psychopathic traits required to engrace (and force) their way into key positions of power). These folks are way too optimistic, or are deliberately trying to end the world.
Wake up. The power is already available to the sociopathic billionaires running Facebook, Google, Twitter with the most advanced AIs. We're going to get more of what they want. I hope OpenAI lives up to its mission of making AI more useful to most people.
Giving AI the ability to improve itself shouldn’t even be a debate.
it is impossible to avoid
@@xonious9031 technically no, but human nature is what it is...
People need to understand that the types of solutions selected from a solution space are a form of communication and whether we understand that communication or not, the selectors and creators of those solutions will.