The idea of infinite memory is compelling. In a way, humans already have infinity memory but not infinite recall. What we are talking about here is infinite recall (or Total Recall). I wonder how much people are thinking about the ramifications, or psychological consequences of infinite recall--remembering too much? The underlying presupposition here is that forgetting is a bug rather than a feature of consciousness.
One of the issues we worry about a great deal when working with highly skilled individuals in critical jobs is Negative Training. Because memory is long term we don’t want to train the individuals to do something incorrectly, then retrain them with the correct methods. Think of a Dr, Astronaut, or Airline Pilot. What if they remember the wrong way to do something because they never forget? AI with infinite memory will never forget the wrong way to address a situation, even if you 👎🏼 the result.
@@manuelbeltranvalle Yes, I have had to deal with this phenomenon in my aviation career. It becomes increasingly difficult with age to unlearn evolving standard operating procedures (SOPs). Maybe an oversimplification, but, in the case of an AI with "infinite memory," I imagine this to be less problematic. I.e.: "Solve the following problem given the following constraints" one of which is: "DO NOT do the following!"
He was absolutely talking about NotebookLM at the 5:00 mark, and i am right there with him. When i started using it, it almost made me get up and quit my job--cause it is *that* good. It feels like it came out of where but it didnt, and itnis as bad as it is ever going to be. Why does the world need to hear my voice, this sounds and feels better :D I also remember where i was when alphago solved go (and previously when chess was solved). The forst one, chess, had me excited. Go on the other hand, made me a bit afraid. Why? Npt because the models were necessarily smarter, but because they could generate better answers based on reasoning we couldnt pissibly understand. Not unlike seeing robots who have invented new ways to "walk". You get this eiry feeling that there is way, way, way more to this universe than even the nuttiest human could imagine :) And to the humans who say, "well it cant even tell me how many Rs are in strawberry" i ask, What are the first 40 prime numbers. They are different problems which take different kinds of intelligence to solve, both are useful, and we cant ignore either lest it be our downfall. Now, where is my flying car and jetpack?
00:06 Steady Progress Over 15 Years 06:00 AI will be an amazing teacher and doctor 07:30 Near infinite memory transforms user experience 09:21AI becomes more accurate and controllable 11:50 AI acts as a judgment-free conversationalist 17:14 Embracing Multidisciplinary Learning 18:30 The Power of Collective Intelligence P/S:If you want this on the right side of the video, install GPT Breeze.
I am currently reading his recent book The Coming Wave in lucid prose, which is informing me about AI and AGI digital technology and gives me cause for concern.
One algorithm changed everything. Speculative Decoding. Simple framework for parallel intelligence. Infinitely scalable. Turned out to be an older method but developed over 15 years.
Hey Everyone 👋 Find the parts that interest you: 0:06 - Continuous improvement over 15 years 1:10 - Prediction as the primary skill of intelligence 6:00 - Memory as a transformative capability 9:21 - Feedback loop enhances AI accuracy 12:14 - AI as a conversational companion 17:17 - Embracing multidisciplinary skills for the future Chat with videos via Bumpups 🌲
They never talk about the manipulation and fakeness of the information to which the search engines and AIs have access to. The talk about you’ll get access to everything… and yes you’ll have it! But ‘everything’ is manipulated and falsified. To get to truth you have have previous knowledge of the subject to have a criterion otherwise you’ll be mal informed.
You are correct. AI censorship - benevolent or not - is the biggest threat to free speech. There will be an abundance of information, but getting to the truth on a matter will be extremely difficult with an AI goalkeeper on the dataset, especially if the AI has been trained that a particular topic is taboo.
I agree that prediction is very important, but what we as a species predict is how the environment will respond to us. ChatGPT does no such thing. It just predicts the most likely next token, based on static knowledge and a prompt. So this is where the AI's have to head imo: true prediction and interaction with the environment.
If you haven't already, read "AI Snake Oil" by Narayanan & Kapoor -they debunk the existing claims of predictive AI. But if you constrain AI to simple physical interactions with the environment then the laws of physics will apply so, for example, "what goes up must come down" becomes a kind of prediction it will be able to make accurately. Yet that result is not a probable outcome but 100% inevitable (well unless an object reaches escape velocity). Some police departments in the U.S. are currenty using predictive AI for precrime assessments on individuals' future predicted criminal behaviour and that is probablistic prediction, not deterministic. Tesla's self-driving AI is making mistakes (sometimes with fatal results) because it cannot make flawless predictions in a probablistic physical enviroment. Given the inevitability of real time edge cases, it is not clear how AI will ever be able to solve this prediction problem (and certainly not with next token prompting based upon static knowledge as you point out). So I am not sure there will ever be "true" prediction coming from AI.
@ Please define how can you be « near » of the infinity? You either are at infinity or you are infinitely far from it. 😂 Just to show that he is using a sales pitch which is not a trustworthy representation of the reality. Please exercise caution and judgment.
Scaling and data is nearing the end for this area of exploration. Time to find more efficient architectures and learning strategies that are similar to nature's methods. Evolution has effectively been reinforcement learning but intelligence and survival as the target variable.
apart from memory arrange her to can listen to music and see videos or movies or documentaries,,, it will be easier for her to analyse instead of asking all the time, takis
Mustafa Suleyman gives me hope; in spite of his rational justifiable evidenced based pessimism for our lived current dystopia experience of our daily lives in Blue Planet Global Village . see his book The Coming Wave .Bring on his ilk!
Science discoveries have stalled. Little real progress in the standard model. Currently technology is based on fundamental discoveries from 50 years ago.
Copilot is bs. ChatGPT is great for general stuff, Claude is a star for coding tasks. MSFT just isn't cutting it. Sorry to say that, but that's the truth
Two years ago, we didn’t have ChatGPT. Now we have rapid advances that are accelerating because of these companies. Unless you, yourself, are solving problems for the masses, it seems like you are biased complaining with no solutions. These companies are doing great work, let’s be appreciative. And each month there are advances. Who knows how far they will get on the behalf of the rest of us that use their technologies.
The problem is that thinking is not what makes humans special. Perspective and observation are what make humans special but because we also think people have been trained to act like machines instead of humans...then when machines can think better than humans they have an existential crisis.
Wow, he's gained weight. Must have been a stressful transition. Good interview. I'm very excited about NotebookLM future development - as the quality and configurability improves it will become a mainstream tool for learning.
Lets race to replace all people, who needs people. I just wonder how people will have anything when everything will be owned by the very few. Our current global system is broken. AI is not the solution and is actually a threat to people. I want opportunity not welfare (UBI).
The idea of infinite memory is compelling. In a way, humans already have infinity memory but not infinite recall. What we are talking about here is infinite recall (or Total Recall). I wonder how much people are thinking about the ramifications, or psychological consequences of infinite recall--remembering too much? The underlying presupposition here is that forgetting is a bug rather than a feature of consciousness.
One of the issues we worry about a great deal when working with highly skilled individuals in critical jobs is Negative Training. Because memory is long term we don’t want to train the individuals to do something incorrectly, then retrain them with the correct methods. Think of a Dr, Astronaut, or Airline Pilot. What if they remember the wrong way to do something because they never forget? AI with infinite memory will never forget the wrong way to address a situation, even if you 👎🏼 the result.
@@manuelbeltranvalle Yes, I have had to deal with this phenomenon in my aviation career. It becomes increasingly difficult with age to unlearn evolving standard operating procedures (SOPs).
Maybe an oversimplification, but, in the case of an AI with "infinite memory," I imagine this to be less problematic. I.e.: "Solve the following problem given the following constraints" one of which is: "DO NOT do the following!"
I have had an extremely good memory for decades now and in hindsight, forgetting is better.
He was absolutely talking about NotebookLM at the 5:00 mark, and i am right there with him. When i started using it, it almost made me get up and quit my job--cause it is *that* good. It feels like it came out of where but it didnt, and itnis as bad as it is ever going to be. Why does the world need to hear my voice, this sounds and feels better :D I also remember where i was when alphago solved go (and previously when chess was solved). The forst one, chess, had me excited. Go on the other hand, made me a bit afraid. Why? Npt because the models were necessarily smarter, but because they could generate better answers based on reasoning we couldnt pissibly understand. Not unlike seeing robots who have invented new ways to "walk". You get this eiry feeling that there is way, way, way more to this universe than even the nuttiest human could imagine :)
And to the humans who say, "well it cant even tell me how many Rs are in strawberry" i ask, What are the first 40 prime numbers. They are different problems which take different kinds of intelligence to solve, both are useful, and we cant ignore either lest it be our downfall.
Now, where is my flying car and jetpack?
Great Interview. Layed back that dude. They coocking good stuff in the background i imagine
00:06 Steady Progress Over 15 Years
06:00 AI will be an amazing teacher and doctor
07:30 Near infinite memory transforms user experience
09:21AI becomes more accurate and controllable
11:50 AI acts as a judgment-free conversationalist
17:14 Embracing Multidisciplinary Learning
18:30 The Power of Collective Intelligence
P/S:If you want this on the right side of the video, install GPT Breeze.
Mustafa expresses the benefits of AI very clearly, well done
I am currently reading his recent book The Coming Wave in lucid prose, which is informing me about AI and AGI digital technology and gives me cause for concern.
wann wird das kommen?
One algorithm changed everything. Speculative Decoding. Simple framework for parallel intelligence. Infinitely scalable. Turned out to be an older method but developed over 15 years.
My wife has selective infinite memory lol
1.45K subscription? This should be 1.45 million channel.
Hey Everyone 👋
Find the parts that interest you:
0:06 - Continuous improvement over 15 years
1:10 - Prediction as the primary skill of intelligence
6:00 - Memory as a transformative capability
9:21 - Feedback loop enhances AI accuracy
12:14 - AI as a conversational companion
17:17 - Embracing multidisciplinary skills for the future
Chat with videos via Bumpups 🌲
When will the memory feature / "infinite" memory be released? This is going to be awesome! :D
🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, This guy is amazing!
We will need not only Huge Data Centers but HUGE *Memory Centers* to store all those threads of 0 & 1.
He said "near infinite memory"... check/listen at the 6:58 point. "Near" makes a lot of difference in that sentence.
Would you say humans have near infinite memory? If so then that should be good enough, for now.
We’re witnessing the rise of AI legends 💪
Incredible interview! Mustafa is a brilliant fellow. I enjoyed reading his book, "The Coming Wave".
More Tech Bro hype about AI
They never talk about the manipulation and fakeness of the information to which the search engines and AIs have access to. The talk about you’ll get access to everything… and yes you’ll have it! But ‘everything’ is manipulated and falsified. To get to truth you have have previous knowledge of the subject to have a criterion otherwise you’ll be mal informed.
You are correct. AI censorship - benevolent or not - is the biggest threat to free speech. There will be an abundance of information, but getting to the truth on a matter will be extremely difficult with an AI goalkeeper on the dataset, especially if the AI has been trained that a particular topic is taboo.
@ agree
It looks like we would become the arms and legs to a machine! This is very worrisome to see people dreaming about this!
I agree that prediction is very important, but what we as a species predict is how the environment will respond to us. ChatGPT does no such thing. It just predicts the most likely next token, based on static knowledge and a prompt. So this is where the AI's have to head imo: true prediction and interaction with the environment.
If you haven't already, read "AI Snake Oil" by Narayanan & Kapoor -they debunk the existing claims of predictive AI. But if you constrain AI to simple physical interactions with the environment then the laws of physics will apply so, for example, "what goes up must come down" becomes a kind of prediction it will be able to make accurately. Yet that result is not a probable outcome but 100% inevitable (well unless an object reaches escape velocity).
Some police departments in the U.S. are currenty using predictive AI for precrime assessments on individuals' future predicted criminal behaviour and that is probablistic prediction, not deterministic. Tesla's self-driving AI is making mistakes (sometimes with fatal results) because it cannot make flawless predictions in a probablistic physical enviroment.
Given the inevitability of real time edge cases, it is not clear how AI will ever be able to solve this prediction problem (and certainly not with next token prompting based upon static knowledge as you point out). So I am not sure there will ever be "true" prediction coming from AI.
10:12 wonder if that will work with animal languages (ie project CETI)
He's basically talking about PI AI.
Infinite… just that word make me understand that that guy is not a scientist but a salesman.
He said "near infinite memory" ... I even double checked the transcript. See/listen at the 6:58 point.
@ Please define how can you be « near » of the infinity? You either are at infinity or you are infinitely far from it. 😂 Just to show that he is using a sales pitch which is not a trustworthy representation of the reality. Please exercise caution and judgment.
Did you know Microsoft has Crystal Láser Memories working already? 4Tb per cm³
Memória é tão importante quanto tudo o mais.
Good one .
Scaling and data is nearing the end for this area of exploration. Time to find more efficient architectures and learning strategies that are similar to nature's methods. Evolution has effectively been reinforcement learning but intelligence and survival as the target variable.
Hyderabad?
Good day. In the physical world, there is nothing like an "infinite" something. It is just a mathematical concept. Thank you.
apart from memory arrange her to can listen to music and see videos or movies or documentaries,,, it will be easier for her to analyse instead of asking all the time,
takis
open minded embrace is king :-D
Mustafa Suleyman gives me hope; in spite of his rational justifiable evidenced based pessimism for our lived current dystopia experience of our daily lives in Blue Planet Global Village . see his book The Coming Wave .Bring on his ilk!
Science discoveries have stalled. Little real progress in the standard model. Currently technology is based on fundamental discoveries from 50 years ago.
😂 and where is the cloud ☁️ is this supposed to happen? Get real! Fry the brains in twenty minutes easy
Copilot is bs. ChatGPT is great for general stuff, Claude is a star for coding tasks. MSFT just isn't cutting it. Sorry to say that, but that's the truth
Fully agree! Cline extension in VSCODE with Anthropic API [newest llm model] and BAM! 🥳
I’m using copilot in vscode but Claude model inside 😂 works best
You don't realise that Microsoft practically controls openai . It is the biggest inverter and azure runs the openai model training
Two years ago, we didn’t have ChatGPT. Now we have rapid advances that are accelerating because of these companies. Unless you, yourself, are solving problems for the masses, it seems like you are biased complaining with no solutions. These companies are doing great work, let’s be appreciative. And each month there are advances. Who knows how far they will get on the behalf of the rest of us that use their technologies.
Yep, they're not competitive at all. Cursor and Windsurf DESTROY copilot
The problem is that thinking is not what makes humans special. Perspective and observation are what make humans special but because we also think people have been trained to act like machines instead of humans...then when machines can think better than humans they have an existential crisis.
❤
Infinite Memory??? malaka!!! ;))
Hype machine. As if a social science dropout would know.
Social Science dropout, really? I didn't know
There are really no limits to earn money for the tech-bros. I've got that.
I'm pretty sure you're a shyster.
Wow, he's gained weight. Must have been a stressful transition. Good interview. I'm very excited about NotebookLM future development - as the quality and configurability improves it will become a mainstream tool for learning.
Oh yea? And with what infinite energy are they going to use it? Haha
😂😂😂🤣🤣
Lets race to replace all people, who needs people. I just wonder how people will have anything when everything will be owned by the very few.
Our current global system is broken. AI is not the solution and is actually a threat to people. I want opportunity not welfare (UBI).
𝐆𝐑𝐄𝐀𝐓 𝐈𝐍𝐓𝐄𝐑𝐕𝐈𝐄𝐖‼️‼️‼️
"Humans are good at predicting" - total BS. LMAO
👎
Co-Pilot is useless and a gimmick
It's not true it has no judgement. In fact, I find it to be very woke.
This is ai cope
This guy has no idea about artificial intelligence. He is nothing more than a public relations person. Please don't give him any kind of attention
What has he said that you found false?
Sorry, but not sorry. I do not believe a single word coming out of a peaceful. Especially in tech matters.
He started quite a few startups. How about you ?
@DavidHurst-g5h good for him.
Immortality live AGI live singularity live but 🔒 🔒 by decentralized AGI need photo id to access decentralized AGI
Need access ping me