AI Researcher Debunks Michio Kaku's Joe Rogan Episode

[@lucidvizion]

The part that doesn't make sense to me is how someone who is an expert in physics could not know that deriving absolute truth from a string of language is unsolved. He was clearly reading prewritten statements, and I think that they sounded like marketing buzz for research grants.

[@maxlarionov]

Finally someone said it. Kaku has no idea what he is talking about the whole episode

[@maximejkb]

Thanks for watching! Definitely an odd episode that had me wondering where Kaku was sourcing his talking points? The point about quantum computers as a solution to hallucination is just something I've literally never seen anywhere in the literature before. But he said it so confidently.

[@gasperstarina9837]

Kaku went on the Event Horizon podcast. I am not Physisist but listen a lot about universe, understand whaat gravitation is, understand how both theories work but at one point one stops,.. So he was talking about String theory/God equation how atm the formula is on one sheet of paper but soon will be inch long equation of everything. When ask how he starts making freaking poem: atom is lowest line is song, neutron is second, neutrino the thirt and when you add quarks and membrane it becomes a Symphony. Idk like he is half theoretical physisist who knows he will never be able to prove even parts of the theory and some mainstream guy all over youtube... Idk he just doesn't make any sense when talking even about the theory he believe in with his life

[@Kerojey]

Kaku is a fraud and only saying populist cr*p

[@Jejrjrjd]

100% Agreed !

[@alihouadef5539]

Given that he has already provided many incorrect information in the field of physics, I would hesitate to label him as an expert in the subject.

[@maximejkb]

Please subscribe to the channel and share this piece with your friends! Let me know if you want more breakdowns like this of AI-related content. And of course, no disrespect to Michio Kaku -- he's an exceptional physicist and a great science communicator, but peer review like this is part of science! Check out my interview with the astrophysicist behind last month's black hole image "created with AI" -- what the headlines got wrong and what we can learn about black holes: kzbin.info/www/bejne/gafHfqafq9ynfdk If you want more you can check out my video on AI Art: kzbin.info/www/bejne/n6XUmIx8ft-ibtU and my follow-up on Automation vs. Tools: kzbin.info/www/bejne/rHqydoCgrMigi7M Thanks everyone for being here!

[@samirkarki192]

Glad that youtube recommended this video to me. Learned a lot , and as someone who would love to learn more about AI and was pleased by the way you explained, subbed to the channel.

[@maximejkb]

Happy you ended up here too -- thanks for the support!

[@rameenana]

Nicely done man. You explained it well without making Kaku look bad. Appreciate your work.

[@maximejkb]

Appreciate it! Thanks so much for being here!

[@tryingtomakesenseoftheverse]

But michio kaaku is wrong about other things

[@alpacamale2909]

I knew this guy was full of sht when he started predicting the future and saying we will have flying cars and stuff. He should be a writer and Marvel and not much else.

[@thekingofthisworld2154]

Holy shit! As a normie that understands zero about computer programming, I had no idea Kaku was totally wrong. I just listened to the video and though “oh that explains it” and went about my day Had I not seen this video, I’d still believe this stuff. Thank you for putting this video together.

[@maximejkb]

Thanks for being here! Super glad to hear it was helpful!! That's always the goal

[@xmathmanx]

he isn't wrong, he just has a different take

[@41-Haiku]

@@xmathmanx A different take is wrong when it is mutually exclusive with established fact, which is the case here.

[@xmathmanx]

@@41-Haiku how about gatekeeping dude? Healthy or just plain old cunty?

[@dexterdrax]

All the sting theorists are turning out to be hacks...

[@Stalliere]

Your content is sincerely great my friend, so glad I found it. As a professional illustrator and concept artist the way you explained why AI is not a tool helped me greatly to put into words why I don't use it to work (cause it does the work in my stead, making everything about it meaningless). Keep it up man, super interesting!

[@soysorray]

I really liked your respectful approach and how you broke down and explained the topics so clearly. Thanks!

[@maximejkb]

Really appreciate that! Thanks for watching!

[@eliastabuteau6335]

120 likes 2000 views WTF this videos is such high quality content, high quality information, and high quality learning im baffled really! I can't imagine for second that you're not going to expand rapidly on youtube with videos as relaxing as this one , keep going man you got a subscriber😗✌️

[@maximejkb]

Thanks so much! Always grateful for however many people end up seeing and enjoying these videos -- appreciate the support!

[@christinetran9973]

Thank you for releasing this video, it clarified some questions I had about that episode!!

[@maximejkb]

Thanks for watching swaggy

[@cmck1777]

AI does not "internalise" anything, beyond saving a state of current weightings of a model. It is an enormous stretch to think of this as internalising something - especially when it so literally hinges on external verification. I imagine if you work with AIs to the point they become black boxes, it must feel very much like something has been internalised, but it's a philosophical nightmare to think about an LLM using such anthropocentric language.

[@Beauty.and.FashionPhotographer]

Correct !

[@user-jy5qm8nc9m]

Can I ask what brand or type of pen and paper were used in this video ? Very aesthetic !

[@maximejkb]

For the training phase diagrams I was just using a WhiteOut pen! Great for writing on any surface.

[@TARRAN_]

Thankyou for your time , knowledge and valued opinion brother - keep going for us.

[@maximejkb]

Really appreciate the support -- thanks for watching!

[@UnknownUser-in1ok]

We love these debunking videos... Nice work lad

[@maximejkb]

Appreciate that!

[@BryceChudomelka]

Well said. It is so annoying to hear people outside the industry explain these things.

[@xmathmanx]

yeah, so frustrating when you can't 100% control every narrative 😁

[@41-Haiku]

@@xmathmanx B could sidestep your uncharitable reading of their comment by replacing "people outside the industry" with "people who are unable to justify their opinions (by citing relevant experts or demonstrating adequate knowledge of the subject matter)" It is essential for people to be able to discover their errors. Humble, high-quality, informed discourse helps. Confident uninformed speculation does not help.

[@willclavel6320]

if you prompt 'AI hallucination' as is to ChatGPT, then it returns "I don't have the capability to hallucinate or generate sensory experiences like hallucinations"... well in fact you re wrong, you hallucinate without even realizing it :)

[@Beauty.and.FashionPhotographer]

Correct....

[@anthonymichaeldurkin6244]

that was brilliant......very enjoyable listening to a professional who is @ the top of the game..would really appreciate you correcting other inflated AI STORIES....Thank you very much...

[@maximejkb]

Appreciate that, thanks for being here! And that's a great idea, I have some projects in the works!

[@gustavvigolo4480]

then explain to mean how Chat-gpt can spit out "Cotler" whos name is acually Kotler? If it did'nt have mistakes

[@jimmy7144]

Kaku admits on film, I have it, that theoretical physicist and astrophysicist are in the business of guessing they’re not correct about anything they just throw a theory out there and people believe it he’s mocking himself

[@gamezswinger]

I've always said, Michio Kaku is KooKoo. LMAO. He also talks about String Theory's 10-11 dimensions as if they were fact. There is no evidence of that many dimensions in reality. Like most narcissists, he's a charismatic speaker, but also full of sh*t.

[@ghost_in_the_machine11]

from a stanford kid (go card!) this was a cool and refreshing vid. thanks for digging into this stuff and for pointing out the nonsense stated by some of our "thought leaders" you touched on some interesting stuff regarding AI hallucination. i hope AI doesn't place a box around human imagination or dull future humans desire to figure out the impossible. I can imagine a world 100 years from now where AI reinforces facts that are wrong or simply only relative

[@maximejkb]

haha go bears! glad you found this video! and I agree, I think hallucination is a super important problem to solve if we want these systems to be net positive.

[@davudfarzullayev7563]

love the jazz hands

[@maximejkb]

If the hands ain't moving the brain ain't working!

[@prabirkalwani]

Earned a sub homie 💖💖

[@maximejkb]

Thanks for being here -- glad you liked it!

[@bide7603]

from me too

[@javiermarti_author]

Great video. Subscribed. Keep it up!

[@maximejkb]

Appreciate that support always! Thanks Javier!

[@nathanrussell8577]

Cool vid, subbed and liked. Glad I found you doing the KZbin rabbithole 😁

[@maximejkb]

Thanks for the support! Glad the KZbin rabbithole brought you here!

[@damirregoc8111]

Michio Quaku.

[@benjaminkemper5876]

For real

[@emicadic]

Very nicely explained! Great job!

[@maximejkb]

Glad you enjoyed -- thanks for watching!

[@jeremybristol4374]

I couldn't sit through that interview. He was just spoutng imaginative speculation. Thanks for posting this video.

[@maximejkb]

Thanks for watching!

[@WBTuckLive]

Really enjoyed this. Keep grinding homie🤝🏻

[@maximejkb]

thanks man!

[@driesvanoosten4417]

This is a huge problem with people like Kaku, Hossenfelder, etc. They think their opinion on everything matters and people seem to agree.

[@tryingtomakesenseoftheverse]

Why do you have to kaaku and hossenfelder in the same bracket. She talks about physics fundementals. And someone is finally fighting against problems in physics academia. And this man is making clear he doesnt know much about anything

[@theronwolf3296]

Language is not understanding, though it looks like understanding. When you learn something, say, perhaps you are studying calculus, there is a point when *suddenly* the concept of derivatives starts to actually make sense. You mind jump from the linguistic explanation on the page to an internal comprehension. It affects your consciousness (I suspect consciousness is actually essential for this, can't say for sure. From that point on as you read more that framework, is built in your mind, primarily non verbally. You can immediately apply this comprehension to extend your knowledge far beyond the words of the books you've absorbed. The CONCEPT allows you to extend it in ways never mentioned by the words in the book. You can start to apply this to your experiences in the physical world. This is much more significant than statistical linguistic manipulation. You no longer need the linguistics of the book. Your thinking has expanded. It is the 'aha' moment. By contrast the linguistic model is much more like a school child writing an essay. They read some material, put it (hopefully) into their own words, and put it out to paper. AI creates a 'complex' output without ever actually understanding what those words ACTUALLY mean... merely their statistical relation to other words.

[@voiceofreason5893]

Very interesting. As a laymen and non-expert tech enthusiast, I did find it quite interesting that Michio was off on his AI comments. Although he's fascinating to listen to and obviously very smart, he clearly does not have a strong understanding of AI. Subscribed. 427 and rising.

[@maximejkb]

Appreciate that! Thanks for taking the time to comment -- and definitely, no disrespect to the man, he's just an expert in other fields!

[@immoronicisdead]

a little to animated on the hand gestures, but overal a fantastic informative video. great job dude

[@djstr0b3]

He has no clue about computers. I remember him talking nonsense about quantum cryptography

[@dinkaboutit4228]

Kaku (username checks out!) wants to talk about AI because his physics are wrong.

[@bide7603]

Thank god you did this, this wound me up sooooo much

[@maximejkb]

Appreciate you watching!

[@zalzalahbuttsaab]

You seem to have a more accurate understanding of how Large Language Models (LLMs) and AI work, based on current knowledge in the field. You correctly describe the training process of LLMs and address the misconception that these models simply splice together information from their training data. You also provide a reasonable argument against Michio Kaku's claim that quantum computing could solve the problem of AI hallucination, explaining that this issue is more related to software than hardware. However, it's important to note that both yourself and Michio Kaku are discussing complex and rapidly evolving fields. While your explanations align more closely with current understanding in AI research, ongoing advancements in AI and quantum computing could lead to new insights and possibilities in the future.

[@maximejkb]

Thanks for being here and for taking the time to comment! I agree, these fields are moving quickly. For my own curiosity lol: was this comment generated with the help of AI?

[@doubleslit9513]

I thought this was a strange topic for Michio to be writing about. I’ve never heard him going deep on AI and the next thing ya’ know he’s writing a GD book on it? 🤔 And, I suppose as a professor and recognized science expert m, this is an easy trap to fall into.

[@maximejkb]

Yup it's hard to blame him to be honest -- it's the hottest science topic in the world right now and he's a science communicator! Probably some expectation for him to have an opinion on it.

[@41-Haiku]

@@maximejkb I think you're right, which is tragic. I think the most important thing for a science communicator to communicate and embody is epistemic humility, which often means saying "I don't know" when something falls outside of one's area of expertise, rather than relying on the authority of one's name and communicating a shallow understanding as fact. It's difficult to trust a teacher who either can't recognize or won't admit their ignorance.

[@41-Haiku]

I like liking Michio Kaku, so if he corrects himself or later communicates about this topic more accurately while alluding to previous misconceptions, my opinion of him would certainly update strongly in the positive direction.

[@maximejkb]

@@41-Haiku Well said, agree completely! Hard to play that dual role of communicator // public figure and scientist.

[@doubleslit9513]

Right. He’s a good guy with an open mind, especially when it comes to the “are we alone?” question. He tends to take more of an Avi Loeb stance than a NDT “little green men” approach.

[@DarwinsChihuahua]

Michio Kaka, more like.

[@VirginiaCorruption-nt8oi]

"and those 2 things are not mutually exclusive" - so then they overlap and thus his AI and physics overlap in your words. "a large enough LLM that performs this task well...." - qualitative and really subjective statement. Has no relevance to actual performance. Prediction on "question and answer" - CLEARLY has NO experience in trying to actually train an AI bot in something simple like ITIL "object/deviation" for tickets (Outlook - mailbox full) - using natural language. I don't have the patience to keep going. You see the point.

[@sunwukong6917]

1:00 mutually inclusive you mean .

[@heluphicclovanass8954]

Phenomenal video, boyo

[@maximejkb]

Thanks so much!

[@Madasin_Paine]

Outstanding. Keep this up. To keep oneself on the bleeding scalpel edge of medicine in the US, I've heard of courses on KZbin, established by it's well known and well respected universities that have relatively breief yet rigorous challenging courses on deep learning that today's medicos must master or become an even drier gold fish stuck to the floor as far as relevance goes. Obsolete Man doesn't appeal to me nor do so called experts expecting the gullible public to assume their expertise should be presumed in unrelated fields. Can you recommend 3 or less such self teaching modules? UCSF allegedly offers one... MD PharmD centered models would be ideal. Thanks in advance for any advice to move to the next step.

[@pqqt1046]

kaku has been making that exact same speech for 10 years now.. every interview is the same

[@maximejkb]

It seems he's been doing the rounds on various shows these past few months making these points. Really not sure where he got this talking point about quantum computing solving hallucination? I've never seen that point made anywhere in the literature...

[@anthonymichaeldurkin6244]

this is regular viewing...im understanding ....tu..Maxime Kawa..

[@Lucas-wq9yl]

I watched the episode and was like I don’t have info to properly disagree but it didn’t sound right.

[@maximejkb]

haha your gut was right! Thanks for being here -- appreciate it!

[@xmathmanx]

when you see the word 'debunked' you should assume it is being used wrongly

[@junbert9720]

this dude is funny hallucinating like crazy he just said problem is not problem hahaha this is joke

[@JasonJrake]

You were way too nice to Kaku. He should be expected to know better than making sweeping statements on a topic that he clearly hasn’t even read introductory level material about.

[@midbc1midbc199]

Ai doesn't exist how most people picture it in their mind.......it's machine learning

[@AndresR-td5jr]

Im bombarded with his content an Neil Tyson, and i came to the conclusion they are the "influencers" of physics who limit the curiosity of people. Every time i see their videos is like the still live in the 1940s. Keeping science stale

[@Draxis32]

Perfectly done. If we could harvest the energy on your hand movements, we would probably be able to hit our carbon targets for 2030. P.S: just a cheeky joke

[@maximejkb]

lolllll those hands do be moving! appreciate you being here!

[@rttp-righttothepoint6656]

hes an expert in public sceintism. Hes not regarded as legit by anyone that matters. Hes just a face for science.

[@HumanBeingSpawn]

One question for you? Have you considered becoming a choir conductor? Your hand movements are very distracting. Maybe cut out your head and share some random videos related to the subject matter?

[@scottymccrae]

Breaking out "Algorithms" 😍

[@maximejkb]

had to had to

[@th3bl1nds4murai]

I think those guys go all universally knowing are probably never challenged by their own kids. Or he doesn’t have a child who makes him look at his own mistakes and apply a moral compass against his assertions. God mode allows you to violate the rules. He’s probably in god mode from his populous books

[@rickshawty]

he is out of control

[@mattmorris419]

Maybe im dumb but this video just sounded like you saying hes wrong and then explaining exactly what he said with more technical jargon.

[@Beauty.and.FashionPhotographer]

Correct...seems You are one of only a few, who actually listened to what his false arguments were. He contradicted himself in main topic "hallucinations"... see in minute 5:12 .....“ now its true that LLM’s can produce factually incorrect statements,… (this is known as hallucinations, which are precisely mentioned by Michio Kaku) ,…. then this young Researcher continues to say: ….”THAT’S A WELL KNOWN PROBLEM IN AI” This is precisely and exactly , Word by Word, what Michio Kaku said in that, and many other, interviews. There You have it…. what i find peculiarly interesting is how many of the commentors below, are persuaded rather by this young mans enthusiasm, and him opening a small paperback-book form his student shelve, and making some hand-drawn sketches on ai Models, rather than those commentators actually listen to what his debunking “arguments” are. It would be fascinating to see this Young man in a lecture with Michio at MIT one day , and see how he will be arguing his claims to Michio , a top leading Physicists on this planet, who does indeed know how ai software does work and especially how hardware works.

[@benjaminkemper5876]

Not at all, it doesn't just copy and paste random crap from its training set, it makes rules according to the training set that effectively become the model. Among many other methods to employ 'reason' and assess how to respond. Michio is a quack, he is so far outside of his field of 'expertise'

[@Beauty.and.FashionPhotographer]

@@benjaminkemper5876 Michio Kaku uses Computer tech and software that Ai coders would die for to have access to. Arguing this is not his field is the comment coming exclusively only from an uneducated person per say. here is a solution for YOU: He still is giving lectures at NYU , You should go to one , sit in, and then raise your hands and tell him exactly this what you just said. He will then listen and ask you questions, you soon will see how much of a kindergarden student you are compared to this man. he is a kind man, and you will find yourself rethinking your own logic about ai,....But thats something someone like You would never dare to do. i can provide You when exactly in which lecture hall he will have his lectures. but You would never ever have the guts to go there and challenge him. for obvious reasons. He would very fast dismantle your arguments,.... as you already know very well. Thats why people like you rather stay anonymous and drop some random stuff here on comments on a youtube video.

[@GRSsix]

Bruh I've seen that guy spit so much BS even about physics how do people take him seriously at all

[@ARSALANKHAN-ub3hm]

Your hand moving is very visible

[@maximejkb]

The more they move, the more I care!

[@ARSALANKHAN-ub3hm]

@@maximejkb but the content was great and straight to the point ...

[@maximejkb]

@@ARSALANKHAN-ub3hm Appreciate it!

[@ediekimo9110]

I can't blame michio because after all AI is not his field of expertise. He's a good astrophysicist that's for sure

[@maximejkb]

Definitely! No disrespect to Michio Kaku. Thanks for being here!

[@LaughingBat]

Seeing him talk about quantum computing... I have trouble trusting anything he says about anything at this point. The problem isn't that he doesn't have any knowledge, but he has demonstrated pretty clearly that he will not be honest when speaking outside his area of expertise. It's sad because he has great energy and his presentation style is great. But you just can't trust anything he's saying.

[@boogieman6529]

Astrophysicist ?didn't know he was one the problem is noone is taking him seriously since 90s when he started doing shows promoting science. Infact many physicists have called him out for spreading lies in public

[@kimblecheat]

Yeah-he just seemed to dribble on in wibbly words about a lot of things this time, Ol Michiu- maybe he’s getting right old

[@rameenana]

He took a big hit with string theory, yeah?

[@davehart9972]

Dr. KuKu

[@tonyli8703]

he is smart guy but used it on all the wrong stuff, for example, string theory

[@Jejrjrjd]

He is not also expert in Physics LOL

[@SuperJoshuaWorld64]

Are you related to Kyle Rittenhouse

[@RicardGomes76]

NP TO YOU

[@DanielTravieso-d3i]

Stop with the flailing hands; it’s driving me crazy

[@NineSeptims]

Thank you kawawawawawa

[@maximejkb]

lol thanks for watching!

[@rbx22]

wtf's wrong with your hands ?

[@ApacheMagic]

As someone who is neither a physicist nor an AI expert, it seems the point still stands. Regardless of how it works, at the end of the day the AI cannot distinguish between what is true and what is popular, most repeated belief or even ideology. The fact is, we can’t vote on the truth. Whatever software you create to resolve this will require the thing understanding this difference and seeking out actual truth, and somehow knowing the difference between the true source of actual data points vs the true source of lies, false beliefs and ideologies. The problem with that is something we all struggle with. Belief stands in the way of truth, every time. Off to ask CHATGPT what a woman is…

[@maximejkb]

Thanks for being here and thanks for watching! I hear you -- I like what you said about the conflation of what is true and what is popular. If certain falsehoods are repeated frequently enough, then a well-trained language model will reflect that: those falsehoods *are* probable continuations. That's part of what makes hallucination such an entrenched and difficult problem. However, the particular mechanism and failure mode Michio Kaku was pointing out -- verbatim repetition based on a mental model of LLMs as sentence-retrievers -- is wrong. Thinking of making a video about current approaches to the hallucination problem. Because there are some kinds of hallucinations that can be addressed -- for example, factual correctness: there's no ambiguity about whether 1+1 = 2, or about who the president of the US in 2010 was. Still an incredibly difficult problem, but less daunting than the really hard task: separating misinformation from information, which we can't even seem to get right with human moderators.

[@95np23]

I think you are a bit out of base here. Kaku was keeping things simple for viewers and to a fair extent the general statements he made were right. You say so yourself. Also, you are an AI researcher, not an quantum computer expert. What he was trying to say is, with quantum computers it will be easy to build something that will be able to be a better fact checker! Why? Because the way quantum computers are build, building a software to compliment that and solve the fact checker problem is going to be hella easy. Kudos on the click bait title. Debunks Kaku's theory 🤣🤣