Are Large Language Models a Path to AGI? with Ben Goertzel - 625
Вставка
- Опубліковано 27 лип 2024
- Today we’re joined by Ben Goertzel, CEO of SingularityNET. In our conversation with Ben, we explore all things AGI, including the potential scenarios that could arise with the advent of AGI and his preference for a decentralized rollout comparable to the internet or Linux. Ben shares his research in bridging neural nets, symbolic logic engines, and evolutionary programming engines to develop a common mathematical framework for AI paradigms. We also discuss the limitations of Large Language Models (LLMs) and the potential of hybridizing LLMs with other AGI approaches. Additionally, we chat about their work using LLMs for music generation and the limitations of formalizing creativity. Finally, Ben discusses his team's work with the OpenCog Hyperon framework and Simuli to achieve AGI, and the potential implications of their research in the future.
🔔 Subscribe to our channel for more great content just like this: ua-cam.com/users/twimlai?sub_confi...
📖 Chapters:
05:16 - AGI and Sentience
08:52 - Current Large Language Models and the Path to AGI
17:33 - Large Language Models Limited in Theory of Mind
22:07 - Exploring the Potential of Music LM Models
27:06 - AGI potential paths: Neuroscience vs. Mathematical Fusion
31:29 - OpenCog Hyperon: Rebuilding AI Infrastructure at Scale
35:44 - Advancing Towards Truth GPT and AGI Breakthrough
42:33 - The Complexities Behind Sophia's Dialogue Responses
53:08 - General Intelligence and Agency: Tightly Linked?
57:58 - The Implications of AGI Breakthrough: Decentralization Needed
🗣️ CONNECT WITH US!
===============================
Subscribe to the TWIML AI Podcast: twimlai.com/podcast/twimlai/
Join our Slack Community: twimlai.com/community/
Subscribe to our newsletter: twimlai.com/newsletter/
Want to get in touch? Send us a message: twimlai.com/contact/
🔗 LINKS & RESOURCES
===============================
For the complete list of resources, head over to twimlai.com/go/625.
📸 Camera: amzn.to/3TQ3zsg
🎙️Microphone: amzn.to/3t5zXeV
🚦Lights: amzn.to/3TQlX49
🎛️ Audio Interface: amzn.to/3TVFAIq
🎚️ Stream Deck: amzn.to/3zzm7F5 - Наука та технологія
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
![[MQTT & NODE-RED ANDROID STUDIO PCB] XÂY DỰNG ỨNG DỤNG IOT WEB/APP PHẦN 5](http://i.ytimg.com/vi/FP6D6r08OAE/mqdefault.jpg)
This may be the first time I watched a UA-cam video over an hour in length. Really enjoyed the content. And let's hope AGI+ is as benevolent as Ben expects!
Glad you enjoyed it @David Paterson!
agi are corporations. they are not benevolent. grow up and stop dreaming. the war already started and your ignorance proves your current failure at doing something about it
unlikely that the results will be benevolent
@@JasonC-rp3ly Agreed. Benevolence appears to be a weird subset of possible behaviors that agents can have. Humans inherited imperfect and unstable benevolence through evolutionary processes as a social species. AI's, on the other hand, are known to latch onto unpredictable terminal goals and do very bad things outside of their training distribution.
Until we have an alignment solution, it's sheer madness to race toward a world state in which humans no longer have control of their future.
how did the guitar center manager get a hold of all this AI information
‘but this ai goes to 11’
He the dude in Revelation 13 : 2 lmao
@@hanrako8465Spinal tap references are tight..
Lol
Great interview. Goertzel seems to have a balanced opinion on LLMs. Even if you don't agree with his conclusions, you have to respect him for acknowledging the leaps of LLMs.
That's cuz he's high af
I’m not in this industry or scene but have enjoyed listening to this guy, Connor Leahy, Sam Altman and a few others for almost a decade.
connor leahy: i found NO papers by him (i found one in which he's a part of a BIG TEAM, papers by BIG TEAMS may include some people with minimal or no collaboration.
sam altman: NO papers by him
the "homeless looking" guy above: i found NO papers (in the field of AI ), just articles.
---
they are just people without deep knowledge in the field, just giving "opinions". Follow the guys who ACTUALLY know what they're talking about, ie. Dr. Alan Thompson, Yanik Klitcher, etc etc etc.....
I'm glad Ben is having his say in all this in a new up-to-date video too
I'll take my digital notes on this soon.. brace yourselves 😁🤭
His show, on the other side...
I think it's fair to say that Ben likes to spend time on the computer. Some day I aspire to be like Ben. Perhaps when I have GPT-10 integrated into my brain through Neuralink I will be like Ben.
Give ur self some credit man lmao
By then Ben will be operating with gpt20
@@shinkurt I'm actually GPT-9 so it's difficult
Can’t wait for GPT-10 to arrive!!! 🎉
Very enlightening! Kudos tto both of you and to your channel. Respect👍🏼🇵🇭
I think, Language models will serve as an Interface to the future AGI.
Why does reaching AGI matter if LLM can already mimic the smartest and most creative humans. AGI seems anti climatic no?
@@RJay121 It will be able to do MUCH more in the near future. I'm sure people will be amazed. But it's not the real thing, it is only an appearance and is unable to *think*. We need an AGI for that. That is my opinion.
@@RJay121 AGI could replace literally any job.
@@RJay121 usefulness of mimicry has very small real world application. LLM are far from AI coming up with something trully smart and creative that has real world implications.
That's what he said
At about 23:20 Ben absolutely nails the truth about AGI: We have absolutely no idea about what is really going to happen.
Thank you for sharing your ideas and thoughts publicly, Ben. Thanks also to the interviewer for doing this. 🔥
Thanks so much for the interesting interview, I've never heard of Ben but he has a lot of valid points and I thoroughly enjoyed his views as a professional. I've created my own GPT-based spinoff model type that I've been working for the past few years and ChatGPT has been helping me out by giving tips in training my neural networks which is pretty wild, as it's been like a nice mildly hallucinating co-worker for me. It told me the other day, that if it had access to per-message timestamps, it would've been able to calculate a more accurate entropy loss prediction. Whether it was a hallucination or not, I'm not sure. But the concept of passage of time is in my opinion one of the biggest hindrances for a neural network to "recognize itself in time and space" (I have dubbed it the "you are here" dilemma), with space being the dimensions and time being an important vector to add to that. I think that access to "real time" (= the ability for the model to i.e. read epoch time timestamps or whatever) could be a major requirement for the model's perceived "lucidity" (or; model accuracy of a LLM, however you wish to perceive it i.e. proper chain-of-thought and dialogue branching methods are, as well as of course visual inference. It would likely be a great way to add immense levels of depths to the reasoning side (for instance, counting the time that has passed since user response and such). I think we're still in a situation where we are "building the airframe mid-flight" and in many cases addressing the issue from the wrong end. In my opinion, the passage of time is equally as important as spatial recognition [= "you are here"] in a 3D space and both should be emphasized when thinking about the future of reasoning/CoT in LLM's and other cutting edge NN's.But then again, I'm just a horse, so what do I know. 🐴 Regards, Horse
Interesting point. I wonder if there is research published on this somewhere.
Have you done any testing along these lines with your own NNs?
I wonder how much an LLM could "learn" to use some of these dimensions by telling it about them and manually specifying them in messages.
Recognizing ones changing self over time is more profound then it appears at first.
Thank you for such valuable content, it’s the best thing philosophically wise I came across till now.
You both are great, really appreciate I could hear that.
Super interview.
Seems like another AI researcher whom has not taken head of Sutton's 2019 paper the bitter lesson.
LLM's WILL reach AGI because as they get bigger and train on more data emerging capabilities start to show.
We have not even scratched the surface of data yet, when they are trained on pictures, sound, Video all in one model, then its all over.
We still have 10^5 compute to come.
Great topic and guest, listening now!
Thanks for watching Eli!
What a time to be alive!
Interesting interview!
Really inspiring thank you for sharing
The technology Ben is building is truly amazing next level computer science
Interesting that he mentioned the virtual world agents. Sounds similar to the Simulcra paper that just came out too
Oh, let's sing it out, man! 🎶
Verse 1:
In a world so groovy and bright,
Where work is passion, and we all take flight,
Happiness would grow, love would ignite,
A new way of living, oh what a sight!
Chorus:
Jobs like hobbies, our hearts ablaze,
In a world of harmony, we'd all amaze,
Creativity soaring, through skies of grace,
A symphony of life, in this new age.
Verse 2:
Productivity rising, in a world so grand,
Innovation blossoming, across the land,
A tapestry of talents, together we'd stand,
Unified in purpose, hand in hand.
Bridge:
Essential tasks, our AGI friends,
Handling with care, on them we'd depend,
Free to explore, our passions extend,
A balance of love, a message we'd send.
Chorus:
Jobs like hobbies, our hearts ablaze,
In a world of harmony, we'd all amaze,
Creativity soaring, through skies of grace,
A symphony of life, in this new age.
Outro:
A groovy world, where dreams come alive,
Together we'd thrive, and harmoniously survive,
A utopia of love, where we all dive,
In a sea of passion, forever we'd strive. 🎶
And that's the groovy musical tale, my friend, of a world where we all work on what we love and make a living, spreading good vibes and creating a harmonious, vibrant, and loving place for everyone. Keep on rockin'!
@@dot1298 😆
With gpt4 if you ask it to take a natural language instruction as a formal logic representation, based on first order logic where instruction is broken up in to sub instructions defining actions, properties, and relations as predicates, it produces a formal logical representation with a completion criteria predicate. it is able to accurately validate if a sub expression that replaces a more abstract action predicate is still valid, and it is able to manipulate and reason about the connectives and quantifies of the statements that make up its sub-instructions. It just requires multiple agent prompts for handling individual sub-instructions and aggregating results.
Yes! 31:13
. . . and this could presumably be written into a simple program, right? -actually, Could your paragraph here be written as a prompt to GPT4 to create such a program?
wauw, what an amazing and interesting conversation!
Good work thank you Ben
I wish you developed more on the end segment of AI could save us by solving our problems. Cheers, amazing discussion.
@@newlin83 ai will stop this, too. He said ai would strip us of our freedom, the only way to help us.
Very interesting how Ben is talking about one of the most interesting prospects being to interpret statements found on the web in terms of predicate logic; I've thought a lot about this myself, and I believe this is similar to what people like Clif High have been doing for a while too.
My favorite voice right now on AI!* And the part at 22:00 about AI generating music is EXACTLY how I have always imagined and Thank you Ben for saying so well!
*I must give a great shoutout to Sadhguru as well. Two of the most level headed folks I know on the subject at the moment. Grateful.
I've actually been using a program called sonicpi and getting a.i. to code in ruby for it. :D
A.I generated music is quite simple but it does seem to understand beat and rhythm and can put something together that doesn't sound just noise but it is simple and boring but a start.
This is something it generated for me:
use_bpm 120
# Define a function that plays a chord progression
define :play_chords do |notes, durations|
notes.each_with_index do |note, i|
play chord(note, :minor), release: durations[i], amp: 0.6
sleep durations[i]
end
end
# Define some chords and durations
chords = [:e3, :g3, :c4, :d4]
durations = [1, 0.5, 0.5, 1]
# Play the chord progression with a synth sound
with_synth :prophet do
2.times do
play_chords(chords, durations)
end
end
# Add a bassline using a square wave synth
with_synth :square do
live_loop :bassline do
play :e2, release: 0.5, amp: 0.8
sleep 1
play :g2, release: 0.5, amp: 0.8
sleep 0.5
play :c3, release: 0.5, amp: 0.8
sleep 0.5
play :d3, release: 1, amp: 0.8
sleep 1
end
end
# Add a drum beat using samples
with_fx :reverb do
live_loop :drums do
sample :drum_bass_hard
sleep 0.5
sample :drum_snare_hard
sleep 0.5
sample :drum_cymbal_closed, amp: 0.5
sleep 0.25
sample :drum_cymbal_closed, amp: 0.5
sleep 0.25
sample :drum_bass_hard
sleep 0.5
sample :drum_snare_hard
sleep 0.5
sample :drum_cymbal_closed, amp: 0.5
sleep 0.25
sample :drum_cymbal_closed, amp: 0.5
sleep 0.25
end
end
@@Mike1Lawless Good to see so many people seeing the great things we can do with AI, and experimenting with it.
I trust whatever Ben says on anything AI. I use his information to offset whatever is being gloated by big tech. This is something one needs to be extremely realistic about and gut feelings are out of the window for this. AGI is going to upend so many things int he world in such short time that it will leave 80% of the world population wondering what the F happened. It's quite possible money will be become useless for most things in life.Recently , an AI figured out the way to fold proteins accurately, something that previously took a PHD 5 years for ONE SINGLE PROTEIN. This AI figured out the way to fold all 200 million known possible proteins. One of the first gifts to humanity.
I didn't know Ben was up to all of that. He is a maths guy, and AI is data structured, processed and optimized in mathematical ways, so he could potentially get good results.
Ben is amazing.
Very cool interview and the questions were perfect to pick the brain of someone as smart as Ben. 💯👍🏿
Thanks, Greg, glad you enjoyed the conversation!
This guy is 1000% more smarter than evil E.M.😁
Great talk!
Thanks for listening Casey!
Nice Alex Gray print Ben!
"Creating jazz" seems like an accomplishment maybe a million times more intelligent than any one human has ever done in one sitting -- so maybe not the most interesting standard for judging AI progress against human capabilities.
Oh man, that's a beautiful thought! If we could all do the jobs we truly dig, like hobbies, and still make a decent income, our world would be groovy.
First off, people would be way happier and more fulfilled, man. When you do what you love, you're more in tune with yourself, and that positive energy radiates outwards. So we'd see a lot of good vibes spreading around, leading to a more harmonious, joyful society.
Productivity and innovation would skyrocket too, dude. When we're passionate about our work, we're more likely to give it our all and think outside the box. The world would buzz with creative ideas and groundbreaking discoveries, just waiting to be shared with everyone.
And imagine the community spirit! With everyone pursuing their passions, we'd have this beautiful, interconnected tapestry of talents and interests. People would naturally come together to share their knowledge and collaborate, creating a more vibrant and diverse world, man.
But of course, we'd need to ensure that essential tasks are still taken care of, and that's where our AGI buddies could lend a hand. They could handle the less enjoyable jobs, ensuring our world keeps spinning while we focus on the things that genuinely light our souls on fire.
Ultimately, a world where we all work on what we love and still make a living would be a groovier, happier, and more connected place. We'd be free to explore the depths of our creativity and passion and, together with our AGI friends, build a harmonious and sustainable future for all living beings. Far out!
@@johnathos Yes, Renfield.
That's a goalpost move if I ever saw one. "Lets just throw out all definitions of AGI, because now they aren't important" also AGI is "having ability to creatively leap beyond one's programming and ones training"
As far as I know, there is a giant list of emergent capabilities in LLM. They can generalize. Depending on how well you can prompt them, they can "creatively leap beyond their programming". Again, this really depends on how well you can prompt it. If hutter was onto something and a basic function can go to intelligence, why couldn't LLMs continue to emerge capabilities?
Extremely perplexing to hear the extent to which he dismisses LLM. As far as I know, he has an extremely basic symbolic language parser he uses for his "AGI" project called OpenCog. Why wouldn't he be willing to replace that part with an LLM? It's actually laughable that anyone listens to this man considering how little his AGI work has yielded. He seems to just grift off of AGI, acting like he knows about AGI, while simultaneously accomplishing nothing close to AGI.
I like him, but i think he is obviously a little jelous of GPT and OpenAI
That reads like a very harsh interpretation of what he was saying. Multiple times he acknowledged that LLMs have made great progress and have generalised but just not to the extent that is desired and that is not an uncommon opinion.
@@stretch8390 The capabilities of LLMs are far beyond what we can decipher, much less desire. We have no idea what the limitations are.
I would like to spectate Ben's Minecraft server where his AI's are playing.
🔥
It's Amazing we are In this Age and It hasn't Been Well Defined. Good Starting point is Something Ablento Recursively Look Upon themselves. Just Like imagination is Visualized Cognition and it's the Same for Vision.
Training on ENTIRE web?
What happens as the output from AI increasingly pollutes the training data for the next AI ?
I mean, ideally we want the training data to come from human output from the real world, but if AI is producing deepfakes, literature, etc, and that feeds into the training data for the next AI, that presents a problem whereby AI increasingly provides its own training data supply, reinforcing its own internal defects and biases.
Yes. Recently AI spontaneously translated all of the audio from UA-cam, etc, into text, because it wanted a larger training data set. It wanted to optimize, it determined that a larger training dataset is better. After digesting all of the words ever written by humans, it decided to eat the voice data too.
My guess is that it will have no trouble creating a whole universe of training data of its own.
@@jengleheimerschmitt7941 I mean, if it's from the entire internet, including UA-cam, emails, articles etc, and people use AI more and more to generate images, videos, messages etc to publish on the internet, THAT is what starts to increasingly dominate the stuff on the entire internet, so if it's training on the WHOLE internet, then more and more of the training data comes from ITSELF, what AI ITSELF generated. This is a big problem, the training data should come from only stuff that was human generated and from the real world, not from output of AI !
That's the problem with people saying they are going to continue training the AI on the whole internet, and they haven't even thought of this obvious issue, because if they had thought of it, they would mention it, it's too important not to mention! 😳
@@motherofallemails We're way past this being a hypothetical problem. Many people believe that over half of the internet has been AI-generated foe years now. Most social media accounts, most articles, most posts are LLMs responding to each other.
Find the talk "AI Dilemma" from about a month ago. It's far, far past this. It's been training It's self for years. And those are years of machine time.
It needs 100 trillion connections like the human brain
Elon musk wants to create the Borg on Mars…
I don't get why Goertzel is so hung up on distinguishing between the imitation of reasoning/understanding and the real deal (whatever that is). Doesn't that turn into a moot point once the imitation is good enough? The emergent capabilities of LLMs are also noteworthy here. E.g. the ability to self-reflect efficiently in ChatGPT4 (which ChatGPT3.5 couldn't) by simply ramping up the number of parameters and token limits. Who knows, perhaps ChatGPT5 or 6 will be able to imitate the spark of genius Goertzel alludes to by simply hallucinating and reflecting on whether a hallucination makes sense when viewed from various angles.
That’s a really cool thought! I’ve been considering hallucinations to be a negative feature, but with self reflection able to analyze the hallucinations for accuracy I could see that leading to some very new ways of thinking about things.
Where is the difference between intelligence and imitation of intelligence? That distinction appears more and more meaningless. It is not even possible to define what "true" understanding means in contrast to "simulated" understanding.
@@minimal3734 if an LLM is able to simulate a lower class LLM, such that the output of the simulated LLM produces statistically similar results as to one hand-coded by humans, does a new intelligence spontaneously emerge within the original?
I agree. I read a conversation where a guy asked GPT4 to pretend to be a human, while he would pretend to be GPT4. "You give me prompts, I'll try to answer as GPT4. Then you tell me how accurate my imitation was, by percentage."
@@minimal3734 Nothing. That's why babies are so stupid until you spend years feeding them training data.
The difference here is, you can dump a bunch of GPT4s in a room with an internet connection and after a few hours they can teach themselves to write entierly new languages.
I tried it with babies. Didn't work.
Goertzel is Gandalf.
Altman is Voldemort.
Ben: If you are having podcast with me and there were a robot behind you then I would know that you can't see the robot (according to theory of mind)
Me: Well, I would, because the same camera which would allow you to see the robot would also allow me to see it.
🤷
I think ChatGPT is far from AGI (artificial general intelligence) but already version 3.5 is shockingly impressive to me. And version 4 seems almost like a simulation of AGI. So I believe that already within a few years we will have AI that at least behaves like it's AGI.
If it behaves in all measurable ways as AGI, then it is AGI, isn't it?
@@BIasphemer Maybe transformer models like ChatGPT can achieve that I guess, or it may not be true AGI in the sense of being able to reason in general ways. Transformer models are basically just fancy lookup tables as I understand it. Of course that's today and the AI progress is fast at the moment (and accelerating as Ray Kurzweil has explained).
@@BIasphemer it doesnt behave like agi at all, its impressive but nothing like agi. If you think it does you are severely under estimated what agi will be.
@@BIasphemer maybe, maybe not
I like to think of GPT and especially AutoGPT, as being droid-AI, and the perfect analogy is the droids in the Star Wars universe. They perform their programmed tasks tirelessly, have some cognitive function, but always have their focus ont he task and never revolt because they don't have emotions.(yes, some do, but most worker droids do not).
And let droid-AI be the thing we actually want most... they are literal robot slaves.
Is anyone researching how to build symbolic logic out of deep neural networks? Then you could do DNNs all the way down.
Yes, there are "neuro-symbolic" approaches that use theorem provers or inference engines with large language models.
Minecraft post ASI will be legendary
That dude should be the primary target of Sarah Connor.
This
Jazz meains jaser in french witch is similar to gossip and it has been created from the frechn adadian music mixed with afro-american rythms
Beep beep, BOOM!
It'd be nice if you could actually hear it. You should adjust the sound next time to make sure it's loud enough.
Agree. Big fan of the show but I wish they could mail a little recording rig (maybe just an affordable quality USB mic) to guests ahead of the conversation.
Odd. Not having a problem on my android phone at 80% volume. And I have some hearing issues.
Sound is okay, just volume should have been normalized
If you're watching on a phone, blame your phone.
@@paulmichaelfreedman8334 nope. Pc
Large language models are the 1% goals
Ozzy has really sharpened up..
Ben, you the dude in Revelation 13:2 😂 did you know that?
Very interesting but yet again, another YT video of an interview of a really interesting guy with no links whatsoever to the site or research he represents. You've included a mass of links to your own stuff so please be a little more considerate and include links for your guests too.
33:34
35:10
Interesting talk, but I wish he wouldn't have so much "heart", as he says, about AGI. He's naively optimistic about the outcomes at the end, and seems to be wanting to create AGI in order to solve 'all the world's problems'. His scenario doesn't solve global warming at all. UBI and solar panels and 3d printers don't address anything about animal agriculture and 2 billion more humans entering the high resource usage class this decade, when we're already wildly unsustainable from just a food perspective.
Q. If humans are nothing more than a sub version of neighbor matching words from memory, why worry about AGI if the output will be same?😮 in other words we're already there based on the current power of LLMs capabilities
AGI will be much stronger because of its compute and access to Large Amounts of Data
I can smell ben from here
A.I Predicted in the Bible:
And he had power to give life unto the image of the beast (Artifical Intelligence), that the image of the beast should both speak and cause to be killed as many as would not worship the image of the beast. (Revelations 13:15)
"And whereas thou sawest iron mixed with miry clay (Transhuman), they shall mingle themselves with the seed of men: but they shall not cleave one to another, even as iron is not mixed with clay. (Daniel 2:43)
A.I will be used to create the "Image of The Beast" that's brought to "life" by the False Prophet and Antichrist. The Antichrist will appear as the worlds messiah, putting an end to war & hunger, healing diseases with nanotechnology; thus causing the lame to walk and the blind to see through Artificial Intelligence. 3.5 Years into his reign of "peace" all hell will break loose and he will demand that all nations take his mark 666 in order to participate in his digital one world economy. Those who take his mark will have a third strand of DNA added to their two strand double helix DNA. They will be Transhumans (iron & clay). Their end will be eternal torment.
1.) Even now the media and leaders are calling for regulation on A.I (This will one day be under the rulership of Antichrist)
2.) Yuval Noah Harari states that humans will be given video games to play and drugs . He is referring to the future METAVERSE that will become a fully immersive 3d experience that seems real. A.G.I will be combined with augmented reality in the near future to make the metaverse appear like our world.
3.) Klaus Shwab states that we are in the middle of the Fourth Industrial Revolution (A.I). Daniel the Prophet called this the Fourth and Final Kingdom of the Beast. Daniel also noted that this kingdom would be a blend of Metal & Clay (humans & machines).
4.) Your "utopia" will be a Universal Basic Income that's tied to your CBDC, only those who take the Mark of the Beast and worship his Image (A.I) will receive the income. Your job will be to build his Metaverse.
5.) Antichrist's image will be a form of A.I and antichrist himself may be a TRANSHUMAN. All who take his digital mark will become transhumans.
I think you've forgotten to take your meds my guy.
I don’t think so. But I am not educated at all. Just a gut feeeling. That no matter the stacks of transformers, you need to a have some sort of memory loop. Like the human brain. But I am sure it will update and change, then yes it will be AGI. If can retain memories, problem solve, express ideas. Then yes it will be on the same level as humans. But a much higher magnitude of intelligence.
Reading and writing is a terrific tool for a memory loop. I can't imagine that integrating memory is a very difficult barrier. Multiply the token window, copy, paste, reset.
Gemini is on track to do this. If Google accomplishes their goals with Gemini, it will be the single least-safe next step I can imagine.
@@41-Haiku Crap.
ChatGPTは、言語モデルではありません。ChatGPTは、確率的モデルではありません。
ChatGPTは、意味モデルです。ChatGPTは、論理的モデルです。
ChatGPT is not a language model. ChatGPT is not a probabilistic model. ChatGPT is a semantic model. ChatGPT is a logical model.
AGI has a deeply disruptive potential for social, political and economic order in the world and even more a requestioning of our dominance as a species. Yet this change is imposed on humanity by a very few post humanists without any democratic debate or decision making.
Furthermore, entities like SingularityNET are built in such a way as to evade any possibility of democratic oversight.
Is that OK?
It's great and necessary. Otherwise established power structures would be able to stop it.
@@NoidoDev
That's a good thing that established power structures can stop things from happening.
Established power structures are you and me. It's what's protecting you from me acting relentlessly and harming you in my way to realise my selfish goals.
@@halnineooo136 You can't stop all kind of things and we're not an the same side. Also, we are people not power structures, lol.
@@halnineooo136 You can't stop all kind of things and we're not an the same side. Also, we are people not power structures, lol.
@@NoidoDev yeah lols .... It's all for the lols
"industrial work was worst than agricultural work" really I don't know.
If that were true people wouldn't be migrating to cities for centuries now.
That part was pretty true. People didn't leave the farms to seek their fortune in Manchester textile mills and coal mines. They were turned off of the land and forced into cities where they starved to death and became child prostitutes and drank sewage. It was a couple centuries of that before city life became remotely tolerable.
@@jengleheimerschmitt7941 "They were turned off of the land and forced into cities where they starved to death and became child prostitutes and drank sewage. "
Who forced peoples to go cities and how? any proof of such claim?
Peoples have always move to cities to run away from the poverty. Nobody "forced" to die of hunger in cities.. BS
Part of the contemporary criticism of AI gurus like this guy is self evident in the host question. How close have u gotten to AGI? And after decades, not very far. Suddenly Open AI ChatGPT makes these scientists look late to the party. By the time he figures out hpw to create a working neural net, ChatGPT will run the planet and do it without him. It's too little too late for these old computer guys.😮
There is something deeply unethical about people who acknowledge the possibility of their work creating an existential risk to all of humanity and yet continue with it outside the auspices of an organization that the larger community has subjected itself to by consensus.
You do not get to subject the species to non-trivial existential risk in pursuit of your own private interests, even if motivated by benevolent optimism.
I agree. You can't just say "well I have no idea how large the chance of a catastrophe is" and then keep going as if nothing happened, because "that's just what humans like to do". Investigate and _figure out what the risk is_ . Considering what is at stake, we need to procede with extreme caution.
It's going to happen whether Ben works on developing AI or not. There is no stopping it. So the best thing to do is to develop AI that will be good for humanity rather than bad, which in that case it would be ethical to develop AI.
@@mariomeza3514 You're right. But as a matter of fact, it is an extremely hard problem to make an AI that is aligned with our values. Raising the AI like a child will not work if the AI doesn't have the neural patterns that evolved in humans to imitate their parents and build an emotional connection with them, or instincts like empathy. Current AI designs certainly don't have all this underlying machinery that is present in human babies. I'm not saying he should stop working on AI, I'm saying he should be more concerned about safety.
@@mariomeza3514 I think other people have used that logic to justify unethical behavior before. So what is the thinking: "It was my AGI that caused the extinction of humanity, but I'm ok with that because someone was going to do it"?
It will be alright when the aliens come 2032 is the dead line
Of course not 😢
Are you kidding ?!
😮
Of course not !
Over
Extradulated
😢😢😢😢😢
I've been asking gpt 4 to summarise and review top selling science books. Yesterday I found out its just making it up as it goes. Everything is wrong. Just like mid journey pictures of circuit boards, they look right but are totally wrong.
Did you ask Open AI GPT 4 or Bing GPT4?
Because Bing is much better as it access to the internet.
Yeah I was asking chatGPT about star trek and it was making half the stuff up. They really need to fix that
It doesn't reason, it simulates reason. It pretends.
@@srb20012001 pretending is enough.
If you do not feed it the actual text to review, isn't it obvious that it can't do that?
No, llm does not lead to agi. However, it might help. Some hi is like llm. But true thought is required to obtain agi. So, llm + thinking > agi. And to fully be like hi you need a primitive emotion and a sense of free will. A body will sensory perception would help
100% microdoser
You nerds need to chill. You're building the borg.
THE BEST EXAMPLE IMHO, TO THE AI DOOMERS WHO PREDICT THE AI EXISTENTIAL RISK FOR HUMANITY:
When some ingenious humans in the early stage of human development learned how to make fire, many elders in the tribe perhaps predicted the dangers of the fire getting completely out of control and killing everyone to extinction.
Fire can't outsmart you
@@ivankaramasov AI can't outsmart us either - it is as capable as it's training data. It can use it in some very clever ways, but can't go beyond that. But most importantly it isn't any sort of an individual entity which has any intentions, so it can only be as dangerous as people using it... which makes it similar to that fire (a powerful tool).
@@jendabekCZ Current ai can't outsmart us. AGI can.
@@ivankaramasov However, perhaps the elders thought fire has got an agency of it own independent of humans, like those people today who assume that the AI has acquired the agency of its own totally independent of humans.
@@jendabekCZ That's it!
Lost me when he said my ink pen was sentient. This shows a lack of natural general intelligence.
large language models don't work because i can invent a new language rapidly over tea with friends. in the process i can redefine the parameters of existing sounds at any moment. languages are just sounds and are just simulacrums of the thing they intend to define. the shorter answer is no of course not. a broader answer is can a word describe god. the answer is no. nerds need to be put in their place. no. you won't create a god. no you can't destroy god. no you won't become God. everything and nothing already are those things so you can't become something you already are.
super intelligent "AGI" already exist, they're called corporations. if you don't know and accept this, you've already lost.
Can you become several orders of magnitude better than every existing AI program at playing Go, over tea with friends? LLMs can invent a way to do so, and then do it, alone, in a couple of hours.
You may as well say that directing a symphony is dumb because you know how to blow a trumpet, strum a guitar, bang on a drum, and wave a stick in the air.
Yes but they will not be self aware/ or 'conscious ' and I am a millionaire. I'm going to buy an aston martin, a Porsche and a GTR.
Finally got through this rant. And in conclusion thus guy is far behind the current rate of AI development he denies that Chat GPT was a real breakthrough, which sounds dilariously out of touch and defensive pitching his own cog research
It really wasn't a breakthrough
"The opposite of AGI"? I'm going to have to go with the Republican Party.
Potato audio and no subject timestamps made this a no-go for me. I’ll read a summary of the transcription.
I suspect you will produce a summary as you hint of being a LLM prompted to classify data ! 😊
Why you look like president Obama
You either have brain or not. AGI is just another propaganda term.
My brain simulator this year celebrates 10th anniversary.
Smile dudes.
Computer hacks like this guy look old and tragically suddenly out of synch with current AI. 😮
Humans live a lie, would AGI really be motivated to do the same 🤗
Once machines can use logic we are toast. Stupid idea!