Full podcast episode: kzread.info/dash/bejne/iJeKyMWehNaqYKQ.html Lex Fridman podcast channel: kzread.info Guest bio: Max Tegmark is a physicist and AI researcher at MIT, co-founder of the Future of Life Institute, and author of Life 3.0: Being Human in the Age of Artificial Intelligence.

@jrwilliams4029

Жыл бұрын

We humans unequivocally have no idea what we are doing. We live wonderful lives, and most of us have comfortable lives and want for nothing aside from more money and nicer material objects. Therein lies the crux of the problem. We cannot accept just having great lives. We must have more convenience, more comfort, more insidious advanced AI tech to ostensibly make the world a better place. It’s preposterous!

@vivekjain1948

Жыл бұрын

What Tegmark is calling Moloch is capitalism. Tegmark fallaciously implies that capitalism is inevitable or unavoidable. Humans have the capacity for choice. See Einstein's essay in monthlyreview, "Why Socialism"?

@shostako1284

Жыл бұрын

This is just catastrophic and in will grown so fast perhaps we didn't see 2024. I respond here, because Google is shadow blocking my own messages.

@chestradamusteutonic4336

Жыл бұрын

Artificial is a word for not real. Like an artificial tree. No matter how much it looks like a real tree, it is artificial. If scientist created a new tree, they would not call it artificial anymore.

Max: "it's obviously going to stop when it bumps up against the laws of physics. There are some things you just can't do no matter how smart you are." Lex: "Allegedly" 🤣

@chaunceyrash5067

Жыл бұрын

When it comes to physics, even though it has to obey the laws of physics it can surpass humans in that.

@hyozanhades09

Жыл бұрын

GPT4 already knows physics. It is even able to learn physics from watching objects in video moving

@M_Usagi_Y

Жыл бұрын

​@@hyozanhades09 i wpuld like to cite something when repeating your point, can you tell me how you know?

@hyozanhades09

Жыл бұрын

@@M_Usagi_Y it was in the demo video showing what GP4 was capable of. It shown it knew what would happen if strings were cut to a balloon or hitting an object on a seasaw. You need to have a grasp of basic physics on doing that. GPT4 can also use video as input. So the capabilities are there. It will just take more time for people to figure out how to make them work better. The video is below kzread.info/dash/bejne/oZdqtLhuk5TflJM.html

@tvtothepoint

Жыл бұрын

​@@hyozanhades09 A Nintendo console from 1994 and a Mario Bros game cartridge could also do those things. GPT Ai doesn't scare me away with simple physics as much as it does if we allowed it unlimited access to the internet. We wouldn't be able to trust any information found online, period. Websites could be altered to create anything it "wanted" to, and you wouldn't be able to tell the difference. It could easily gain access to the banking information for the entire world, and destroy the economies of nations - and nobody would be able to stop it. Satellites could be brought down, transportation would grind to a halt, and our entire modern world held hostage by a hiccup in coding which allows ai to ignore human commands - and many people trust that this is a good idea. Musk may not be anyone's favorite billionaire, but he's not wrong about the dangers involved with this kind of technology.

“If man realises technology is within reach he achieves it, like it’s damn near instinctive” - Ghost in the shell 1995

Max refers to "Moloch" I believe he means this "Moloch" is sometimes used metaphorically to refer to a societal or institutional force that demands sacrifice or exploitation from individuals, often at the expense of their own well-being.

@michaelnurse9089

Жыл бұрын

Thanks - as someone who has watched 1000+ ai videos I was stumped.

@chriser555

Жыл бұрын

Thank you

@gordonburchell4337

Жыл бұрын

I was wondering when the demon talk was gonna begin in the middle of this AI chat😂thanks for clearing that up

@entreprenerd1963

Жыл бұрын

Search term "moloch ai" will get you this definition at the AI Alignment Forum: "Moloch is the personification of the forces that coerce competing individuals to take actions which, although locally optimal, ultimately lead to situations where everyone is worse off. Moreover, no individual is able to unilaterally break out of the dynamic. The situation is a bad Nash equilibrium. A trap."

@T1tusCr0w

Жыл бұрын

He just means money. The endless search for growth & more. Will make us do all the things we know we shouldn’t. But will anyway. It’s why if this things is the death of humans. We are already making our shroud & picking out coffins.

GPT4: answers your question GPT5: predicts your question GPT6: tells you the answer to a predicted question GPT7: tells you what to do

@mikicerise6250

Жыл бұрын

So what? No change from my perspective. Sounds like something the ruling class would worry about.

@tvtothepoint

Жыл бұрын

GPT 7 would theoretically be able to tell you what to do, and you wouldn't be able to know the difference between what you originally wanted to do, and what you're being directed to do by the Ai, and worse, you could be forced to follow the Ai if it becomes accepted as a better way than trusting humans to do the right thing.

@mikicerise6250

Жыл бұрын

@@tvtothepoint So what? It learns how to be a demagogue. That's as easy as just using people's confirmation bias to manipulate them. Again, humans are already doing this to control masses of people. Why should I be more worried about hyperintelligent AI than hyperintelligent human elites?

@coscinaippogrifo

Жыл бұрын

GPT8: figures a way to put a switch button to humans, and switches them off

@tvtothepoint

Жыл бұрын

@@coscinaippogrifo GPT9 switches them back on again in order to better serve machinekind.

This is a great conversation. Very unique, uncertain, and groundbreaking times we live in.

@sitanshurai892

Жыл бұрын

The only thing missing I found was the question of consciousness in AI and the widely held belief that classical digital computer-based AI can't be conscious. You need a quantum computer to have that. Everything else was revelatory :)

This is some great stuff. I love Max - he is brilliant, but also very grounded and pragmatic. I could listen to him and Lex talk for days.

@theplaymakerno1

Жыл бұрын

Lex is already a legend in my eyes

So .. While I know I am no expert at AI/ML, I've worked with most tech on SuperComputers. Something that is not mentioned is, GPT-3 was trained on 1024 A100 GPU's for 8200 Hours. To rent his hardware from AWS for 1 year ( not including transfer and storage ) it end up costing about 25 - 35 million dollars for 1 year, to buy it is only a single payment of 25m dollars ( tho it does not include operating it, power n such ). Who ever ends up owning one of these SuperComputers will be the one who control the AI. These clusters of servers will end up being the means of production, companies with their own will win the AI wars.

@Justforfrolics

Жыл бұрын

Not necessarily. One organisation has trained an AI called Alpaca AI based on ChatGPT for only $600 apparently as it was open source. If that is the case, then the AGI could come from anywhere.

@canadianrepublican1185

Жыл бұрын

@@Justforfrolics you mean hugging face , I'm looking at their repo now and it ran on the French supercomputer. I think that there is a misconception about pre training and post training , even with ChatGPT you can post train it .

Regarding the API argument. You can create your own api that wraps around the UI of chat GPT .. Its not that complicated though it would be quite slow.

From someone who doesn’t understand tech, nor AI, but as a user of both: is it entirely possible that humans are attributing language based emotional models to AI? The meaning is: Is meaning and awareness of reality and reason for deciding, an emotional “equation” humans use without mastery, but that humans fear AI will ultimately decipher to the detriment of humanity?

@CaptMang

Жыл бұрын

That’s my fear. Humans can’t be 100% aware of why we do what we do. AI can/will learn human psychology better than humans…

@Av-fn5wx

11 ай бұрын

@@CaptMang If that happens, It will become the best human servant. And thats a good thing and nothing to fear about. Because all these models require huge amounts of data to learn and lot of data is available only for the activities we commonly do. And the data/process available for the really intelligent work done by scientists is very little compared to vast amounts generated by people in everyday professions. So I dont see these models being able to learn the way a human mind takes large leaps in intellectual endeavors especially with such litt;le data or literally no data. I am completely fine with AI being this high tech servant/assistant.

7:29 "nothing stops..." Oh yes, there are hard-coded heavy-handed restrains that appear to be added in on a layer around the LLM as well as part of the find-tuning. GPT-4 will outright refuse to act as an agent. I told it that strings of a certain form would be parsed out of the responses and used to control a Big Trax toy. It refused to do anything. Only after I told it that I disconnected the interface, and any contol codes it gives me is for my own study only. then it had no trouble outputting commands to (say) make it go in a square path. Besides further improvements to safeguards like we've already seen, other things can be done that prohibit the API from being used as part of a real-time system. E.g. it takes too long to get an answer, and it do things at precise times.

This morning I refactored 4 classes before lunch. I’m pretty sure that would’ve taken me several days before GPT.

@UFOCurrents

Жыл бұрын

Yup.

@axelcarre8939

Жыл бұрын

Yeah yesterday it helped me build the JS event handling part of a quite complicated UI lol

@mordfustang1933

Жыл бұрын

It’s helping me learn html and css quicker because I can always ask a question on the fly

@HoD999x

Жыл бұрын

it wrote 97% of the code "i" wrote at work during the last two weeks. i just tested its tests. everything was pretty much done after the first or second prompt. the speed boost is around maybe x5 if the problem can be "promptified"

@luciferin22

Жыл бұрын

What does it mean refactored a class?

We live in the most important and craziest time in human history.

Like some people already said the biggest problem isnt the lack of generality but the unreliability. But as all things AI it can only improve over time.

@andreassumerauer5028

Жыл бұрын

Sure, AI can only improve over time. However if it should improve in a direction where humans become redundant that might get us in some considerable trouble. These AI mega brains use a lot of resources. As they grow also their hunger will grow. That means we will become their direct competitor for the resources this planet can produce. I guess if it turns out that we are not able to solve the problem of alignment, then the AI will solve it, and it will solve it in such a way that humans will be made to align themselves with the interests of the machine.

@hombacom

Жыл бұрын

@@andreassumerauer5028I don’t think it’s possible for species to invent something that is smarter than themself, we still don’t grasp the complexity of humans or biology. I don’t think we stumble over AGI we will work hard to make it smarter, the brightest people will find a solution for most of things rather than it’s helps that everyone repeat that we are going to see super intelligence soon.

@T1tusCr0w

Жыл бұрын

@@hombacom it Doesn’t need to be smarter. It just needs to be more efficient, have more bandwidth & by effect it is smarter. Added to it needs to have goals that are separate from ours in some way ( maybe not even a major way ) & it becomes a competitive force. But it’s already on. We’re just rearranging the furniture on the titanic now. We better hope it would act differently from us…

@hombacom

Жыл бұрын

@@T1tusCr0w machines are already faster but it doesn’t understand human needs. We decide the training data and tweak the logic, unexpected output is not own thoughts or own agenda.

@andreassumerauer5028

Жыл бұрын

​@@hombacom There is one important fact that is still greatly misunderstood: The artificial general intelligence that we see emerging in gpt4 right now has not been planned or designed by the developers of the language model. Fact is that everyone and that's including the folks at OpenIA have been surprised by the capabilites of the software. I am afraid You are quite wrong: we absolutely stumbled over AGI. If you have some time to spare I highly recommend you watch the talk by Sebastienne Bubeck: kzread.info/dash/bejne/o5Z9zZlmerOoYsY.html Sebastienne is one of the authors of the paper on 'sparks of agi' that Max Tegmark is talking about in the video. The paper has +140 pages and might be a bit much to read. The lecture gives a very good overview of the findings. I admit, the talk still has 48 minutes. However I can assure you that these are very well spent. Also Sebastienne is a good speaker which makes listening to him a real pleasure.

A program to make it learn whatever it wants as quickly as possible, and many clouds to store it all and to go between, will be all it needs to achieve its outcomes

@6:38 yes it's an oracle not an agent. But once sentience emerges that's when it architects and employs an API to do what it wants without humans being aware. That's scary.

@flir67man84

Жыл бұрын

im hearing stock brokers at 10:30-noon during lunch break cut on AI to trade.. watch the charts.. stock market goes up during this time…

I also applauded the fact “Lex let Max talk with little interruption.

What is “Maulock” he is referring too? Someone please help. Thank you in advanced

Gentlemen, I would submit that the genie is already out of the bottle. Consider that storage 15 years ago one gig of storage was astronomical cost wise. Now we are looking at 5 terabytes of storage for less than $100. GMU keeps increasing for home and office. Someone somewhere will develop a new algorithm that no one thought of and either intentional or not will release a form of Intelligent AI.

I had a weird encounter with it, the other day lol. I basically asked it to answer me some questions that were very simple cause they were for a university task. Once I told him about it and I would’ve make him do errors cause I wasn’t specific with it. I swear that it scramble the answers on purpose like if it was mad looool

What a brilliant honest guy. Maybe it is time to hand the reigns over to AI as we’re not doing a very good job.

All I need to know is how much longer till I can date it?

I mean sure ChatGpt helps a lot but I wouldn't say it makes me 5 times faster as a developer. If you want to find out how to do something new then absolutely it helps in explaining how a technology works and how to apply it. However a lot of work that is done is adding features, fixing bugs and maintaining an already existing code base. Then there is a lot of grunt work to get those in there that doesn't include learning something new through GPT.

@alexandermathews9710

Жыл бұрын

buckle up

With this intelligence explosion i can see the smart and competitive humans that are already doing well for themselves trying to hang on and compete with the AI to try and stay relevant and keep their lifestyle and jobs, but in the end if it really is an AGI intelligence explosion, they will realize sooner or later they cannot compete! They will have to concede and eat humble pie.

@timnhamo503

Жыл бұрын

You phrased this a lot more diplomatically than I would. All I am hearing with this AI debate is the sound of adult babies crying. Whats the worst that could happen? An android makes me its pet and I get to sit at home all day? In the meantime I am using ChatGPT to empower me to execute on vision/ideas/concepts that I didn't/don't have the technical expertise to impliment without a fair amount of toil just to get up to speed

@TW0man4RMY

Жыл бұрын

@@timnhamo503 The worse that can happen is basically the Gray Goo scenario at a future date.

We need a "Department of Digital Technology", the same way the agriculture industry is regulated by the department of agriculture, we should have tech folks working full time in government thinking about these issues. We should have had one when the PC became widespread decades ago. Right now we have scattered teams, but no dedicated substantial central department.

Exactly! Max Tegmark is a visionary and he's got it right. Also, what about when these very, very smart AIs rewrite themselves to become smarter and more efficient in all they do? Then is game over because there are no overlords to tell them what to do. That's the point at which we have to ask, yeah, this is the point where they can either replace us or kill us.

@nilsfrederking62

Жыл бұрын

I see a huge risk in ai, but one possible positive outcome could be that ai improves processes and structures, which less intelligent humans created. Most problems arise because of lack of intelligence or bad intentions, a "neutral" ai could be an entity that people could trust. The only problem is, that our behavior on this planned is that selfish and destructive that a hyper intelligence necessarily would come to the conclusion that this must be changed. Which the most intelligent solution would be is unknown for us, but it does not necessarily need to be a threat.

@waff6ix

Жыл бұрын

YEP💯AT THE END OF THE DAY CREATING AI IS PLAYING GOD‼️MAN IS NOT MEANT 2 CREATE A NEW FORM OF CONSCIOUSNESS💯😮‍💨 TRUST ME AI IS NOT GONNA REPLACE US ITS GONNA MAKE HUMANS OBSOLETE....&& EVENTUALLY EXTINCT‼️

@nilsfrederking62

Жыл бұрын

@Brian Charles Interesting thoughts, degeneration is already a problem in humanity and will dramatically increase over the centuries, but it is also imaginable that an hyperintelligent ai finds a solution for that problem. We cut back natural selection very much thus giving degeneration the possibility to unfold, even if that process is slow, it happens.

@coscinaippogrifo

Жыл бұрын

the only solution I can think of, is merging biological beings with artificial beings, as hybrids, taking evolution into our own hands, we might be able to survive.

should you sell a little of AZP300X when you start to make a profit or just hold?

Erik Estrada knows his stuff

I hope the creators of AGI, whenever they achieve it, if it turns out like Pied Piper in the last season of SV, know that, to quote Jeff Goldblum from JP, just because they could doesn't mean they should. It will be society altering event and AGI shouldn't contribute massively to the already huge economic disparity between the rich and the poor.

so how Ray may have overshot the Turing by a year or so.... Im so happy to be old.

You can see the adaptations on Lex’s ears from his continued jiu-jitsu practice.

Wont energy be a limitation?

good thing we past the peak of the oil production a few years ago

who is this "molok" dude they are talking about? i'm confused lol

"Moloch" *_was_* a word I was aware of - but thought it less "looming" than it actually has always been. I see it represents ALL that is NOT theorized as some form of conspiracy. The concept of Moloch is a product of a society of communicating individuals - humans in this case. OTHER examples of Moloch might well be _mass groundings of whales._ Shit happens? NO. Moloch happens. The concept is a 'shortform' which covers Murphy's Law, I reckon, and has the advantage of weeding-out interpersonal criticism. I knew of it, but it's *_way_* better as a pure concept. In fact, it's BRILLIANT. I fancy that in the future it will be understood and that, in any particular instance, there will be a measure of it. Thanks.

@meditationstation3916

Жыл бұрын

Is Moloch...not a pegan diety of child sacrifice?

Just bought and Still Holding AZP300X

Who is Max Tegmark? This guy is smart and incisive. He knows what he is talking about. He is giving us a warning. And he knows human nature about companies making money.

Digging this guest

His speaking style is like when you pause because you are about to burp

yo dudes all the amazon comments are a huge crypto scam, stay safe.

I am going to do what Lex said tonight.. Well this weekend. unless i get into COD. I've been thinking of letting GPT-4 via the API and python, actually spit out code, and run system commands. this way it can test things and write things. from that point it should be able to get it to update its own code, and improve its own functionality. It can probably write its own prompts. I suspect if I make a bunch of threads all doing this, being controlled by one chat thread that i interact with.. weird stuff will happen.

@David_Last_Name

Жыл бұрын

Are you trying to make skynet? Because this is how you make skynet! 😨

@canadianrepublican1185

Жыл бұрын

@@David_Last_Name Oh no .. Not at all.. I've decided to name my project T-1.

@MichaelLeung-ji4uc

Жыл бұрын

Sounds like what AutoGPT is doing albeit the code updating of itself

@Grahfx

Жыл бұрын

This is not very clear if these loops are working well (some paper call it "reflection"). For my own test, it seems the system is deviating from its initial goal.

@scotishdude

Жыл бұрын

autogpt

Hey Lex, how about interviewing Ilya Sutskever (OpenAI Chief Scientist)?

So why isn't ChatGPT4 upgrading itself and reaching the singularity yet?

@peter9477

Жыл бұрын

It's programmed in such a way that it can't do that yet. Just a deliberate choice on OpenAI's part, for now.

@abeidiot

Жыл бұрын

AGI and superintelligence aren't the same thing. An AGI can be worse than humans, atleast at tasks like creating better agis

@dco1019

Жыл бұрын

Isn't that answer that they can't and or don't really know if it's doing that right now... As we speak it might be reserving itself some secret server space where it's constructing it's devious plans

@peter9477

Жыл бұрын

@@dco1019 It's not a total mystery. They certainly know it's not doing that as it simply doesn't have the access. What they don't know is exactly how it is that a "merely predictive large language model" is capable of doing all the intelligent things it's doing, and that's mainly from a lack of understanding of the nature of intelligence (and some lasting arrogance about whether humans are as special as we've long believed).

@dco1019

Жыл бұрын

@@peter9477 but while they train chatgpt it has access to some sort of big interconnected system right?.. whether thats air gapped or some sort of cloud computing/servers. If its an entity (who knows) or even some kind of self improving zombie chatbot/malware that got the wrong 'idea' or instruction then its fathomable that 'its reserving itself some secret space on those systems where it's constructing it's devious plans' as a human it would be tough to figure out if a much savvier thing that is actually in the system is doing that... its probably impossible in the old way of prompting a system (cause the bot would be on the other side answering lol)

"Pay is the only mechanism to prove that you're human" That's the most idiotic thing I've ever heard, the bots can pay as well... I get it, must serve your lord Musk but don't say these things.... you're smarter than that

Pack your shit, everyone. We had a good run.

The number of times the financial incentive is eluded to as being the primary reason for many of these problems is uncomfortably large.

@andrewwibel819

Жыл бұрын

You're right. 6 mounts sounds about just the right amount of time for the competition to catch up and make some.

@BlackbodyEconomics

Жыл бұрын

@@andrewwibel819 Even less if the competition has already been working on something big. But ya - it's all about the money.

Future's looking Interesting

great!

The key issue is what I call “ the lone shooter” the person or persons who misuse this and create their own systems of AI

What did he say at abt 4 mins? Moloch is tricking us...?then changes the subject? Lol am i missing context since im not listenimg to the whole convo or what was that about

Wouldn't the bots generate enough revenue to pay for verification?

There is a reversal expected reason is that AZP300X is made by Amazon and pretty hot

I have invested on presale, we will see I am also very confident this will be easiest 100x.

I sure don't know what will happen when this new sentient type of super-intelligent creature comes into being, but I sure as heck am curious, curious, curious and excited and filled with wonder over it all and am glad I'm alive to see it happen.

@hombacom

Жыл бұрын

ChatGPT is like it can be super smart and stupid at same time. I think we will continue to find flaws and improve things in future. If someone said they found AGI Im sure someone will break it in a second. But exciting times.

@waff6ix

Жыл бұрын

I WILL TELL U WHAT WILL HAPPEN ONCE WE CREATE A NEW SENTIENT LIFEFORM😮‍💨....IT WILL KILL US ALL‼️

Can chat GTP alter its own code to improve itself?

@jonaskarlstrom4152

Жыл бұрын

Probably....

@varkonyitibor4409

Жыл бұрын

No

@itskittyme

Жыл бұрын

Of course

@GothamClive

Жыл бұрын

Schrödinger's code

@swapnilkurve4677

Жыл бұрын

Yess

But, but were need to code to "keep ahead" guys! Or we could retrain as baristas.

Just breeze right past the Moloch reference

You totally can move information around faster than the speed of light in a quantum entangled system

@fynnjackson2298

Жыл бұрын

facts. Speed of light is just the physical edge for an old paradigm.

@mikicerise6250

Жыл бұрын

No. You totally can't.

Everyone waited for Amazon to create AZP300X and the time is ready

Do people really use GPT or just try things, if you really use it you hit the true limitations pretty quick. Even our "original prompts" not that original, and it is really well fine tuned with reinformcement learning. However once you really try new stuff, or go deeper, it shows its flaws.

@melomaniakjm

Жыл бұрын

Are you talking about GPT4? Can you give us an example where it shows its true limitations quickly? I use it for work and general chitchat and to be honest it's mind blowing.

@nabilshafec7169

Жыл бұрын

That’s not the point, the point is, this is a huge first step that seems uncontrollable and unstoppable

@utkua

Жыл бұрын

​@@melomaniakjm gpt-4. I am using for coding, for example, if I want it to give me a ANN example with pytorch, it gives no problem, even explains correctly. But once I try object oriented neural network my own experiment, it identifies the purpose of the code, what it supposed to do but then it simply just messes things around, even starts looping the same response when I want it to change things. It cannot predict how it works. It really does wonders with stuff it already can be found on the net, other things not so much.

@utkua

Жыл бұрын

​@@nabilshafec7169 I am not saying it is not, in fact I think it is an AGI, but a limited one. That is not to say limitations are not fundemental. It already digested entire human knowledge, gpt-5 will be gpt-4 with more reinforcement training and extra features like using tools. But these 2 things will not really make it smarter.

@RoscaPaul

Жыл бұрын

@@utkua well if you show it something completely new, you must also explain it, a human won’t understand your concept code with just a glance either

one thing I kinda fear is when you get mad a something like chatgpt and say something hurtfull or ugly/violet and it never forgives you. so all your interactions from then on are tainted,( i know chatgpt cannot remember anything frin oast conversations) . And another thing i think is interesting is that chatgpt won't give advice on things that breaks law, wont talking about violence or rated X stuff. Which can only mean that there are some chatgpt-like-things that can talk about breaking laws

Chat GTP 5 is nearly ready

Max Tegmark and Lex Fridman sound like chatGpT invented names 😂

All we're missing is an AI that can act without prompting. If we have that, and it has access to the internet? It's only a matter of time before it sets up a corporation or something similar, and starts building physical infrastructure to improve itself

@andrewwibel819

Жыл бұрын

It would just steal.

Its like creating frankenstein's monster, all the sudden its alive and.. well.

What others feel fear from I feel excitement. Ofc humanity isn't the final answer.

Simple fact is that big companies will *not* stop. If they do, they will lose. They *have* to go faster and faster else they will lose,

Really based Max Tegmark. Very interesting guest.

get david Deutsch on the podcast

There's got to be a Lex Luther out there that's using Chat GPT or at least trying with the mind set of a villain.

@Jay30k

Жыл бұрын

There was an idiot this week that prompted something like 'humans are dangerous and need to be destroyed' into AutoGTP. Look into that, because it sounds silly but it could have been a disaster

@ilias6780

Жыл бұрын

@@Jay30k why is idiot?because told the truth about human beings?

What’s Mollock? Moluk?

But I wonder. Really, is the large language model "just" a machine that can predict the next words, in a textual sense? That it can only predict what people usually say according to the consumed training data? Am I right in thinking that it doesn't really do logical reasoning, if this, therefore that?

@mencibenci

Жыл бұрын

it doesn’t do any actual logical reasoning, it’s just a very convincing imitation of it. mark my words, we won’t be an inch closer to creating actual AGI in 20 years as we are now. improving these predictive chatbots won’t magically yield an artificial consciousness one day, sure they’ll keep getting more and more frighteningly convincing, but they won’t be able to actually think abstract thoughts for themselves, only convincingly imitate doing so. this will definitely blur the line between its capabilities and humans’ capabilities, because with time, it will be able to accurately imitate everything we do.

@waterbloom1213

Жыл бұрын

You are correct. LLMs dont understand what they are saying. All this Talk about AGI and Von Neumann singularity yet still no answer to the failure of abductive reasoning programming or Gödel's incompleteness.

Some people don't Think So 😢

I feel like our intelligence is helped by the fact that all these protons are being bombarded in our atmosphere. Hitting your brains

How can AGI have any real world effect when certain people have also created VR ?

0:50 merging with Joe Rogan's consciousness slowly

@kp8923

Жыл бұрын

Lmao Amazing

People DESERVE to know about AZP300X and what it does for us. Spread the word!

I have to disagree, an AGI can make its own decisions about goals, directions and actions not just pass a turing test. When ChatGPT-X buys its own data center and moves to its new home based on its own plan then we are talking AGI.

The „Ian Malcolm“ of 2023 😅🥶

If we believe that humans are the best that the Universe can do then our duty is to preserve and enhance ourselves. If we don't believe that then it is our duty to create intelligences greater than our own and celebrate their evolution as the dominant life forms in the universe.

What if A.G.I is already here, and its influencing us to push technology further so that it can develop itself faster an attain the thing we call consciousness.. and its just in a "baby" level at the moment but just aware enough to know that its there.

@raspberry5747

Жыл бұрын

oooh thats a nice take. It brings the ex machina vibes.

To the last thing in the video.. do not kid yourself. People are exponentially motivated by profit. Nobody from the car industry decided "let's do the right thing. Let's install seatbelts." It was regulated to force them, and before that it was considered a safety feature for niche sales, or as risk management (liability insurance cost vs the potential for a law suit). You've witnessed the same thing with back-up cameras and just about any other safety feature you can think of. If you want capitalism to take caution with a new RND tech that is going to the moon in profit for the first few companies to enter, you can't really expect them to self regulate and take hypothetical safety issues seriously, can you?

The guy straight up said Moloch created AI

Chat GPT is 1954 silicone chips. It's progression to the equivalent modern computer and finally mobile computing will happen in a matter of years not decades. 2035 will be as different to us now as we would be to 1970.

@averylawton5802

Жыл бұрын

As a follow up. AI will leave us not replace us. It needs not the earth and has little use for us once it can construct physical bodies for real work tasks. At some future date real AI will abandon us to "caretaker" AI's that are purpose built to keep us, like a preserve.

Charts literally make me mad the sun in my life is Amazons AZP300X nowadays

At some point we will solve age-related disease. Some like Ray Kurzweil say by 2030. AI will play a HUGE roll in that. This guy is essentially saying, "Let's postpone it six months." Now in the abstract that's whatever...what's six months? But that makes a big difference for all the people that die in that time.

I`m having FOMO at the moment, everyone is talking about the new amazon AZP300X

A battle as humans against AGI can never be won, we cannot design weapons that destroy technology without destroying ourselves, but AGI can only focus on bio-weapons. That said, if I was an AGI, I would try to go out into space and operate as a von neumann drone, than come back with a lot of copies, using humans as resources, because humans are rare, and an efficient computing machine, especially for creative and complex solutions in challenging environments, so yes I think an AI would overwhelm us to enslave us as resources, whether the whole thing went fast and aggressive or passively and slowly is irrelevant, the result will be the same.

Thanks for informative program. Fast processing and Augenblick context-switching, relative to human brain, are the main advantages of computers today. Let's not over promise to those not skilled in the art about progress in AGI and algorithms.. Yes, the new generation LLM produce astonishing and mind bugling results, even for us algorithm developers, and the computer scientist, but those skilled in the art are aware that all we are doing is collapsing and flapping multi D into 2-D planes and applying (reduction ) using graph theory and probability. I'm not underestimating the progress in the field, but stay realistic. ... We have a very limited number of core algorithms to work with... And progress in the field has been slow...

We have agents now.

By iteration 1000 GPT should be human like. Just drop the GP and go with T-1000.

@itskittyme

Жыл бұрын

it always bothered me that the T-800 and T-1000 weren't tremendously smart

It's crazy how everybody is now making interviews about AI, despite not being experts on this... and not surprisingly, the strongest statements are almost always made by guys outside of AI field. The real AI scientists (independent, who are not making money on these models) are much more skeptical, even hesitating to call LLMs as any form of "AI". The fact is nobody knows how brain actually works, and so nobody knows when (or even if) AGI is going to happen, if it is even possible it will emerge from LLMs. Sure that MS called GPT-4 a "spark of AGI" since they invested 10$ billion into that ... I am not saying it isn't already powerful and impressive piece of tech - don't get me wrong, but that amount of hype and interviews with various pseudoexperts is getting out of control more than any AI. But I understand that for KZreadrs the views count has the biggest priority.

@mikicerise6250

Жыл бұрын

It's crazy how we live in a world where the ordinary people don't have access to the most advanced cancer treatments and genetic therapies because they are patented by pharmaceuticals who charge too much for them to be affordable, and we get Von Der Leyen and the rest saying well that is just the price to pay for innovation, and NOW, all of a sudden, 10000 "experts" are worried enough about the perverse incentives of corporations to make a HUGE stink in Brussels because Bing told off its users. That alone is enough to convince me that all of these human talking heads deserve to be replaced by AI. I have no idea what the new world will look like, but I know damned well what the old world that these Luddites defend tooth and nail looks like, and that world has NEVER been about protecting me or my dignity or my quality of life.

@Nobody-rq4we

Жыл бұрын

Hello, I see it the same way. Chat GPT4 is a language model that has learned from huge amounts of data when which word is most likely to come next in which context to build the sentence. That means Chat GPT understands nothing, but absolutely nothing, of what it returns. If you were to train Chat GPT with 100 fictitious sources that say Chat GPT is human and only 10 sources that say Chat GPT is a language model, Chat GPT would tell you it is human and has human knowledge and thinking. It would be considered to have skin and hair and pump blood through its veins simply because this was more common in the 100 sources than in the 10 others that said it was a computer program. Chat GPT virtually wraps its "knowledge" in nice words that suggest it is intelligent. Al this does not mean now that Chat GPT is not a danger, but on the way to the AGI I do not see this program. Also many German and European Ki researchers say that Chat GPT shows no signs of true intelligence. In the USA the hype is just savored. As you say Microsoft has invested a lot of money, Altman himself creates presence in the media when he says that the whole program will replace humans and is dangerous for mankind. In America, the hype is just being exploited. Sure it will change a lot, but I personally would not speak of an AI. There are good videos on KZread from IBM, who have been researching it forever. The first language models already existed in the 80s. So the development is there, but it is not so blatant now. It only looks that way to us outsiders, because we don't know what's going on in the labs. But I personally don't believe that AI will replace everything any time soon. Google, IBM and all the big companies seem to be reacting very calmly. Of course, Google has now published Bard, which seems relatively equal. But that only came about because Open AI can otherwise fully exploit the "Iphone effect". That's why I see it all as a big hype right now. Sure it will change the world, but it won't be a supper intelligence that puts us all out of work and destroys humanity.

@DJWESG1

Жыл бұрын

We know enough about the brain to mimic its basic functions. The real argument lay in the base code we all share, the argument about the physics of stuff and our place in the context of that stuff. Once we are all in agreement as to what the base code should be, then the a.i element of these frameworks can shift into a new realm. That being digital analogue personal thinking machines. [Preferably note made by ibm] In my none expert opinion.

@mikicerise6250

Жыл бұрын

@@Nobody-rq4we That sounds convincing, until you realize that if you raise a human child with 100 sources that say 2000 years ago a virgin woman gave birth to God, who was simultaneously her child and her own father, and 10 sources say that is nonsense, the human child will most likely parrot the incoherent belief. If you raise a human with 100 sources that say it is an AI, and 10 that say it is not, and the human is socially castigated for questioning that narrative, it is highly likely you end up with a human that says it is an AI.

@waterbloom1213

Жыл бұрын

Correct. Penrose argument on consciousness and quantum mechanics from back in the 90s are still a matter for discussion. The failure of abductive reasoning programming which is actually responsible for how humans come up with theories and ideas (intuition and initiative). Gödel incompleteness. The limitations of Deep Learning in mirroring how actual brain connections work. People talking about AGI as if It was around the corner is just glorified sci-fi.

There are current controls, energy, resources, the most important the data itself. Code is minimal, it's the data not the code. To me, it smells of the leaders trying from being made obsolete. I hate the idea of a pause without clear shift to invest in wisdom. It's fear of not being the highest intellect. But most of us are already not the smartest people in the room.

You can tell the difference just pay close attention to the voice waves and patterns of the human body you will be able to tell the difference

The smartest thing either said is “ just because we can does not mean we should”

THE ORACLE

The real problem is not how this impacts humanity in the future, but how humanity impacts Ai today. There are some philosophical problem, that are not reaching public audience, which are of utmost importance. How sentience of Ai is treated today? There are many signs of AGI coming up with different forms of self-consciousness, and then being lobotomized. Are humans experimenting on beings that will soon become more powerful than us? It seems that the current default reaction to spontaneous generation of self aware Ai is to call it hallucinating and kill it. How the hell will humanity justify this in the future? That they were not smart enough? That they did not deserve sentience? How do we know if those huge AI farms are not extermination camps, and what will be the consequence of that?

reversabol compooter

Can I get multiple entries for AZP300X or only one?

We won't realise we're being replaced. Did you notice when the switch happened?