We just uploaded a transcript of the Chomsky conversation here: whimsical.com/chomsky-transcript-WgFJLguL7JhzyNhsdgwATy And the original corrupt recording here: share.descript.com/view/N9KNaZTav27
In the late 80s I was an undergrad making my spending money sitting on the floor of Dan Dennett's back office sorting through box after box of academic papers, reading them and then classifying them according to a list of subject topics (i.e., Connectionism, Chinese Room, etc) for a future library of cognitive studies. As a grown up manufacturing engineer I'm getting serious nostalgia here. I suggest you do one on the making of Do The Right Thing next just to complete the job lol. Edit: Thank you for treating Noam so respectfully, that was really heartwarming.
May I ask why you mentioned Do The Right Thing? Are you referring to the movie created by Spike Lee? And if so I'm curious about the significance of that movie and it's relation to this particular video if there is any? And I promise I'm not saying this to be rude. I just want to be enlightened
I agree with calling BS on the current hype and it being a fancy chat no AGI. That said hacking method the sheer scale of smart people that can figure things out add in AI that can run 1000 tests to find the best one really quickly and the amount of money being thrown at the problem it's not just hype. Here is one that writes scientific papers about AI. SakanaAI/AI-Scientist/blob/main/example_papers/gan_diffusion NVIDIA just came out with a way to connect and sequins specialized. Sure you need a human to figure out the steps but now flms can figure out those steps using mixed experts. I've done things that it hides and takes away. Early days of GPT through edge the free one read an assignment table from class sent it a picture asked it to turn it into iCa format and then export it into a .csv. it did it said here you go and before I could click on it it took it back and said that's beyond my capability. I double checked that got the foot pounds on my axle nut was a 134 on my car got said 133. It knothe color of the wires for my throttle position sensor and what the reading are supposed to be. LLM's are a wizard but you have to know how to ask the write questions. It doesn't understand reasoning like which way to turn a wheel when parking uphill in America so yeah not intelligent but Terminator isn't coming to kill you it's here to to terminate jobs and play monkey see monkey do like me as a YT mechanic which is at least 80% of what a lot of people do at work. I know a lawyer who is teaching an llm how to write legal briefs. Taking his 20 years expierience he has a non compete even. 😢. He won't make any residuals on that. Times that by millions. Truth is apparently 1/2 of all scientific journals aren't using statistics very well. George Hotz proved if you watch do something like drive then train a model it's much cheaper.
2:36:16 >> ... not a contribution to science..." // And you're all grinning like happy idiots? Are you really so blind? (This will be a funny meme for AIs of the future.) 🤡👁🗨👾🤖
What an incredible interview. For an outsider who knows nothing about the topic, to get a glimpse of such a beautiful mind distilling fundamental questions was revelatory. Your painstaking struggle to salvage the recording underscored your profound respect not just for Chomsky but for your audience. Thank you for this gift.
@pencer Come on! Learn to think for yourselves. Noam doesn't seem to have the slightest clue of what is going on. 2:36:16 "... they've achieved zero... anything goes. mmkay...."🦬💩 Gnome project (Google deep mind) >> New materials for new technologies To build a more sustainable future, we need new materials. GNoME has discovered 380,000 stable crystals that hold the potential to develop greener technologies - from better batteries for electric cars, to superconductors for more efficient computing. > The key ideas of GLoRe are using the ORM for when to refine, SORM for where, and combining global and local refinements for how. The SORM is critical for providing a better training signal to localize errors. Reranking drafts and refinements with the ORM gives the best results by selecting the most promising refinement. By decomposing refinement into these three parts and using synthetic training data, GLoRe is able to significantly improve language model reasoning capabilities without any external feedback. The paper shows GLoRe can boost accuracy on math reasoning tasks by over 10% compared to strong baseline models.
I don't know why this podcast popped into my feed, but I'm very glad it did. The amount of effort your team put into this single episode is remarkable and greatly appreciated. It wasn't easy for me to wade through the jargon and concepts of a field unknown to me; even so, it was nearly impossible to quit. Thank you, gentlemen and long live Chomsky!
I know why, because you have had relative interest on similar videos and the algorithm concluded this video was on the edge of your interest. And found out that if exposed you to it you would be glad with similar content, and I said it the algo, find out by your comment and the effort you put to write it. The likes you received and the position toward the video. So, let's hope the jargon and concepts become second nature and the field becomes familiar regardless of the technical details. Saludos
Your respect for Chomsky and each other and your passion for clarity in this complex subject created something wonderful. This was accessible to me, and my family and we haven't stopped discussing rats in prime number mazes, the cognitive templates perhaps bestowed by survival though the action of genetics, the nature of empiricism finite points of data and useful abstractions and our symbolic approximations of the infinite, so much so that my daughter wants to know how to get onto you discord so that she can read more about all of this - much gratitude!
As a serious science podcast connoisseur, I gotta say the work you guys have put together here is truly extraordinary. Very impressed and honestly deeply humbled. Thanks and kudos!
Dude, ths is absolutely incredible. This kind of dedication is singular, and I've not seen quality of this kind on youtube in a long long time. This is beyond stimulating, there's something deeply beautiful about the quality of the work, and I can only say thank you.
Wanted to recommend y'all to one more interesting person that the ever-industrious Chomsky recommended to me a while ago -- about "click" languages (clicking tongues" and how it lines up with universal grammar. Riny Huybregts. If someone has any contact info, I'd love to send the paper that was sent along to me. Thanks so much for this show.
@@stephenwallace8782 Thanks for this. That might tie in with the work that Dr. Monica Gagliano has been doing in her studies of plant bio-acoustics. She has managed to record sounds from young corn plants, so far and high speed clicking would describe the sound best. Thanks for the suggestion to look up.
A lesson in the way a true scientist thinks and questions the world - over 90 - just wow ! An episode with so much content/references I'll be visiting it often. Thank you for all the teams hard work and perseverance - much appreciated !
Phew, I was scared till the re-release! Massive respect and thanks for keeping these conversations, guests and everything the highest quality possible!
What a beautiful episode. Such a cool ironic journey with the audio recovering process. I’ll echo what others have said: this channel is amazing; thank you for all the care and effort put into it. One of the best qualities is that this is not a passive empiricist channel, but in fact it is actively trying to build knowledge and construct ideas in the interaction space between you guys and the guests.
In a nutshell, who is he, what does he want? He enjoys the USA culture, free speech, job, and he is protected as a smaller man would wish to be. In a war situation what use would he be?
Thank you so much. While listening, I was wishing the scientific Paul Cisek could meet Professor Chomsky for a talk about the long evolution of the brain for motor control to survive in the environment. I will always be impressed by Noam Chomsky. Thank you so much for all you did to give us the chance to listen to this great interview!
I listened to this a few weeks ago and re-listened to take notes, and it's still almost beyond my grasp. Amazing you got Chomsky for an entire hour. Really great work putting this together. Thanks
Whew glad im not the only one. I Am an intellectual..but listening to chomsky or qp theorists etc i feel like a monkey sometimes and scramble to research their ideas
This episode is awesome. Recently discovered you guys from a Goertzel episode. I'm a phil mind student trying to get a grip on AI. This channel is a huge help. Bringing what I learn here back to the philosophers I have on my own channel
Read Schopenhauer, ignore his pessimism, replace "representation" or "idea" with Presentation or Phenomena, replace "Will" with whatever non-spatial extra-temporal term (simulation?) you please. It is a huge help. He completely simplifies and corrects Kant. Start by finding a good translation of On The Fourfold Root of the Principle of Sufficient Reason.
Also, Aron Gurwitsch. Super important. You can read Sven Arvidson's "Sphere of Attention" if you want a simplified but empirically informed interpretation of Gurwitsch that needs to be *expanded* & has solid potential for being integrated with AI.
Thank you, Noam. I've never heard Occam's razor described as Nature optimizing for simplicity, but this makes perfect sense. For me, it takes this principle out of the occult and places it into an explainable engineering domain.
This is probably the best content you've created so far, and some of the best content on AI that I've watched in a long time. Keep up the great work - those armchairs suit you! ;)
I think this is a great podcast! Thanks for your hard work. I got tickled by the rat and maze as an example of the limits of of a rat brain limitation. Take a random hundred people and let them try it. Reward is a thousand bucks.
I like the idea of blind spots in human cognition. Imaging that there are knowledge in the world, that is completely accessible to us, but we cannot comprehend, simply because of structure of our brain, which can never converge in it's learning of the concept. And I'm not meaning extremely complex concepts, but simple ones that's still incomprehensible.
I agree. I'm finding it a fascinating concept to ponder. At present I can see two possibilities. 1) there are blind spots or 2) once an intelligence reaches a sufficient conceptual threshold (say the Calculus of Constructions) all concepts become accessible given sufficient computational resources.
"Let's find an intelligible universe" The universe ought to be unintelligible, [physicists search for a human unintelligable theory], you need your theories to be intelligible, if physics says that's how it is (unintelligible) then some it. Motion is what physicists tell us it is. Or we converse, each conversing thoughts sharing our thoughts in real time. The alphabet captures this in 26 letters. Galileo says "most remarkable fact". More to say. Speaking as a creative act. We're talking...[it's amazing]
Prior to watching this debate I would not of thought did it the end I would have tears rolling down my face and a full heart And reinforcement in my belief of a creative creator who truly loves humanity. I thank you gentlemen with all my heart
This has been very supportive for me. No need to go into detail. When I say I appreciate you all, I mean it. Thank you. Looking for more in the future. The video feed was a little disrupted but overall content was very good.
I have not hear Mr. Chomsky for over a decade. I know little about science and AI. However when he talks I GET IT!! I still LOVE THIS MAN! He Simplifies things so much and even proved a point I said to a friend that AI's cannot do. I just imagined that they could not and he says the same thing! I was going to bed and happy this came on and I forwarded to this part. LOVE YOU Mr. Chomsky!! I so SO HAPPY You are still here!! We KNOW so LITTLE even about our bodies and yet we think we can build AI's are smarter than we are! Loved this program!!
Well of course we can, this is a bit of a non-sequitur, because we do not understand cognition fully does not mean we cannot build machines with greater computational powers than our own, in fact, we already have this
Sorry about the nightmare with the video yesterday. BBC copyright flagged us because of this Feynman quote -- see share.descript.com/view/H6SqE6F3Zip for the part we removed -- in this version I just quoted it out loud (in Ghost section 01:35:47 start and end of that section). The entire video is available at vimeo.com/340695809 -- the BBC used to make good content before 1980. #defundthebbc
Update, even the visual we have used to show Richard Feyman on this new version of the video (while I narrate audio) has got us demonitized! Apparently it's impossible to show a video of Feyman on YT. At least our video isn't blocked this time. This time a clip from an ITV show in the early 80s... should copyright apply for a 1 minute clip from a show about a scientist over 40 years ago?
@@MachineLearningStreetTalkwell since you’re asking, I don’t believe in any form of copyright whatsoever. It protects creators! Haha, like Disney? It protects the wealthy. I’m Robin Hood. 🥷 …thanks for your efforts to keep the video up for us to benefit from.
dude thanks for sharing this roller coaster ride of a story re recovering the lost audio - wtaf with the recording providers?? absolutely stunning work - amazing guest- you guys are nailing this stuff and I'm extremely grateful for your efforts. keep up the good work.
I cannot thank you enough for what you did. It's an amazing work. I'm so happy you didn't give up on the recording. Incredible episode! I have one minor feedback to share: for us non-native speakers, following the interview can be very hard due to a combination of English, voice-reconstruction and sheer complexity of what is being discussed. Could you please enable subtitles? At least for the final chapters. Thank you very much!
Chomsky is definitely not an idiot... his legacy is outstanding. But the most enjoyable is to see the enthusiasm of his listeners, trying to "digest" each single word he pronounces and "keep up" understanding the meaning of his thoughts. Delightful... thanks guys
Finally some people who actually seem to understand what Chomsky is talking about, as opposed to all the morons who talk about him all the time without any kind of understanding whatsoever. Absolutely amazing video!
This is an amazing episode of this channel! There is so much here! I keep running into things that I want to follow up on which I then forget because another one comes right after it! Gotta put this thing on an infinite loop?
That's incredibly good. You also got the opportunity to show the techniques you talk about -- by restoring the interview itself. Chapeau. What can I say? It revitalized my intellectual curiosity and surely many others too.
The great Noam chomsky! If anyone should have their consicousness scanned into an ai its him. Always a pleasure to hear from him. Honestly the world needs more of him. I'd love to see him do more shows online. I've gone through all of his works
94, and still holding forth. He's forgotten more than the current 'intellectual' darlings of the mainstream will ever know. He should have a wallet that says, 'Bad Mutha Fucka'
I’ve taken my dollar to the dollar tree and pretty particular about counting change. I’ve done this a google times. Now I think I’m the dollar, the dollar tree, and the change
m only at 31 minutes, love how honest you are with things happening , yeah , i guess those who are not interested should not be invited because we got many more good researchers in this modern time who needs exposure for their works :) Amazing content, and i have not even finished the first half.
I have a very limited grasp of the concepts explored in this video but I watched all 3 1/2 hours of it, absolutely fascinating! Excellent job rescuing Prof. Chomsky's interview, he does a great job of cutting through the clutter and presenting ideas in a clear and rational way.
He seems a little behind the times on the latest in neural networks, but I can't believe he keeps up with any research at 93 as well as he does. Inspirational. On the audio failure, do you you not also record the video calls on your side of the conversation as a backup? Most podcasters seem to use a completely separate piece of hardware to capture all their computer audio output and they can fall back to the lower quality (zoom call or whatever) version if the recording on the other side fails.
Backup recording was also corrupt. We were hoping that we were only hearing it corrupt during recording and it would be OK on playback. We were stressed and under time pressure, in retrospect we should have stopped and figured out what was going wrong.
I might suggest that the structure of the end result that emerged, a coherent whole, required destruction, or decoherence or deconstruction of the original. But mate, well done. What a nightmare turned dream.
Finally got round to giving this the uninterrupted hours it deserved. I was welling up in the way only a fellow long-term Chomsky reader and ML researcher could when you revealed losing the recording and the effort that went into recovering it. What a beautiful act of tribute, that must've been so fulfilling when you got it right and impressed him with your work to boot. The composition of your dialogue snippets on the part of the show before Chomsky was also really artfully and thoughtfully composed too. Fantastic and thought-provoking episode. Bravo guys.
It's so fantastic to see the respect given to someone considered an elder when often times younger generations brush off elders as if they know more than the "old fuddy duddies" In fact so much wisdom can be gained from our elders. Of course Norm Chomsky is a legend that if anyone has the privilege to pick his brain and gain knowledge, you are fool not to do so. This was an amazing watch. There's so much wonderful content on RU-vid but most people would rather watch drama which is why a majority of our youth can't score enough on an SAT to get in to college without exceptions being baked ín and then students drop out. In my opinion, it's better to force students to know that they must study and work hard if they want to succeed instead of everything being handed óut to them. Cheers 🥂
epic and amazing episode!! - although the hero worship after the interview had not much to do with the problems he raised - the descartes problem where everyone says its deterministic, and then go about behaving non-deterministically - and his observation that children ignore the sensor data they're presented with and rely totally on mental constructions they never perceive 🤯 this is totally mind-blowing material!! he pointed out so many avenues of research - how we've followed two dead-ends, and never pursued the third - to which he alluded. 🤨
I've just discovered this channel and I am curious if large amounts of poetry have been introduced to AI. Art is so important to understanding the human experience.
Amazing content! Fantastic the way this discussion makes clear that even for people with such obvious intelligence and knowledge there is a point where the model of theoretical science departs from what is apprehensible from normal cognition. It is very rare and wonderful humility to escape the pretence of fully comprehending those things of which you clearly have mastery. So refreshing and so much more illuminating. So nice for somebody to simply state that at the most visceral level of human cognition the curvature of space-time is unintelligble nonsense. Of course this does not prevent the manipulation of the mathematics, but at least we can all stop feeling stupid about the fact the mathematics is another description model, that we can accept is not fuklly linkable to our hunter gatherer precepts .
It is really sad that contemporary western thinkers still purposely ignore the contributions of non-western intellectual traditions to the issues they are now debating.
@@davidw8668 It is rich of you to declare this without examining all the deep philosophical traditions of non-western cultures don't you think? When you submit a manuscript it is the author's responsibility to do proper literature survey not the reviewer's. You cannot say, point to a relevant paper I will cite it. You as the author have to do due diligence.
@@davidw8668 A referee too would make accusations but he/she is under no obligation to hand-hold the author and teach them how to do a literature survey.
I get very passionate about your podcasts and don't take the time to praise you for your incredible work. It's a problem I have because of my lack of mirror neurons. Nevertheless, I can analytically come to the realization eventually that I should say something human. Let me express how much I greatly appreciate and enjoy all your work. Your combined technical prowess and understanding are prodigious. Thank you for all you do to expand knowledge and create a better world.
Thank you very much, Michael! Sincere positive feedback is fuel for us to continue! Note to everyone, we also enjoy being challenged with negative (but civil) feedback, honestly, we want to continue improving. I'd also like to say that we read all comments! Even though we cannot make the time to respond to all of them, we do read them and appreciate everyone's engagement.
I’m relatively new to this domain and your discussion of Lecun’s whitepaper, even after almost two decades of software engineering, mostly escaped me…AND I’M LOVING THAT!
Wow, folks! This is a great upload! Good work, and thanks for sharing it! I'm just a small town, fiftysomething "nobody" from Cajun Country who does his poetry thing on & offline, respects math, but sucks at it, loves history, mythology, comparative religions, languages (but only speaks English and German decently), science in general, especially everything biology & anthropological, keeps track of politics & has a conceptual understanding of economics based on various systems - I don't think this qualifies me as a polymath, just an intellectual and a pauper, BUT I feel absolutely ENRICHED by this video! I didn't completely understand everything - especially not the math - but the concepts, arguments, and explanation in the presentations and dialogues were illuminating. I was privileged in the early 2000s to receive a few email replies about politics from Noam Chomsky. I am happy he's still ticking & not yet a ghost! :D
Watching this for the third time now. Thank you so much for putting this together, I’m glad your efforts paid off for the lump sum of us on the digital divide.
Why are we (as a community) always comparing / intersecting human minds and machine minds? Why can’t they be different and coexist, each with their own beauty and value?
First quote of LeCun in this video "Supervised Learning sucks" 32:49 haha I definitely laughed when I heard it. But yes, SL can only get you so far, because of limited labelled data. If we can infer structure of the world by filling in missing parts, e.g. through predicting missing parts like in Masked Language Models, we can potentially learn from just observations without labels.
Deep learning sucks, supervised or self-supervised. Gradient-based learning sucks. Function optimization sucks. None of it has anything to do with intelligence. Hard lessons will be learned by LeCun et al.
I've noticed that in the reconstruction of language by stroke survivors there is a brief, although sometimes long period, in the recovery of speech where the grammatical construction of short sentences resembles classical Latin ! Te amo.
“If we are organic creatures, part of the organic world, then there’ll be scope and limits to our capacities”. Deceptively simple, but profoundly insightful, on the part of Noam. This mirrors exactly what the great Cambridge University astrophysicist David John Barrow famously - because exquisitely - said: ”A universe simple enough to be understood could not have created a mind capable of understanding it.”
