Taming Silicon Valley - Prof. Gary Marcus

[@MachineLearningStreetTalk]

REFS: - Gary Marcus - Moral Decline of Silicon Valley chapter - amzn.to/4e7jnRC - Shoshana Zuboff - Surveillance Capitalism concept - amzn.to/3ZqHAxS - Alfred Harmsworth, 1st Viscount Northcliffe - "Man bites dog" phrase - en.wikipedia.org/wiki/Man_bites_dog - Evelina Leivada, Gary Marcus3 , Fritz Günther4 , Elliot Murphy5 - LLMs understanding human language - arxiv.org/pdf/2308.00109 - Gottlob Frege - Compositional semantics - plato.stanford.edu/entries/compositionality/ - Jean Piaget - Object permanence theory - en.wikipedia.org/wiki/Object_permanence - Alan Turing - The Turing Test - academic.oup.com/mind/article/LIX/236/433/986238 - Nicholas Carlini - Chess with LLMs - nicholas.carlini.com/writing/2023/chess-llm.html - Mathieu Acher - GPT-4 chess analysis - blog.mathieuacher.com/ChessWinning7MovesGPT/ - Fidelity Electronics - 1979 chess computer - www.chess.com/forum/view/chess-equipment/fidelity-chess-challenger-voice-sensory-chess-computer - Jiexin Wang - AI-assisted coding security - arxiv.org/pdf/2407.02395v1 - Sam Altman - AI-run businesses prediction - fortune.com/2024/02/04/sam-altman-one-person-unicorn-silicon-valley-founder-myth/ - Rodney Brooks - Human supervision in AI - rodneybrooks.com/rodney-brooks-three-laws-of-artificial-intelligence/ - Cruise (GM subsidiary) - Teleoperation staff revelation - www.nytimes.com/2023/11/03/technology/cruise-general-motors-self-driving-cars.html - California State Legislature - SB1047 AI regulation - apcp.assembly.ca.gov/system/files/2024-06/sb-1047-wiener-apcp-analysis_0.pdf - Gary Marcus - Open letter on SB-1047 - garymarcus.substack.com/p/an-open-letter-to-fei-fei-li-concerning - European Commission - EU AI Act - digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai - OpenAI - UK House of Lords testimony - committees.parliament.uk/writtenevidence/126981/pdf/ - A&M Records, Inc. v. Napster, Inc. - Napster case - en.wikipedia.org/wiki/A%26M_Records,_Inc._v._Napster,_Inc. - Jaron Lanier - Twitter Poisoning article - www.nytimes.com/2022/11/11/opinion/trump-musk-kanye-twitter.html - Adam Curtis - Hypernormalization - www.imdb.com/title/tt6156350/ - Russian propaganda model - en.wikipedia.org/wiki/Firehose_of_falsehood - Chris Lu et al - AI scientist paper - arxiv.org/abs/2408.06292 - United States Congress - Equal Employment Opportunity Act - www.eeoc.gov/statutes/title-vii-civil-rights-act-1964 - Luciano Floridi - Digital Ethics research - kzbin.info/www/bejne/j32xeKmsnNZmm8k - U.S. Government - EEOC limitations - www.eeoc.gov/ - Gary Marcus et al. - GPT-4 bioweapons study critique - garymarcus.substack.com/p/when-looked-at-carefully-openais - KAL (Kevin Kallaugher) - AI development paradox cartoon - www.economist.com/the-world-this-week/2023/11/02/kals-cartoon - Sayash Kapoor and Arvind Narayanan - AI Snake Oil project - www.aisnakeoil.com/ - Gary Marcus - p(doom) concept - garymarcus.substack.com/p/d28 - Ezra Klein - Gary Marcus AI podcast - podcasts.apple.com/us/podcast/a-skeptical-take-on-the-a-i-revolution/id1548604447?i=1000592835492 - Seymour Papert - Logo programming language - el.media.mit.edu/logo-foundation/what_is_logo/logo_primer.html - Isaac Asimov - Asimov's Laws - amzn.to/3XTIwtl - Beff Jezos - Physics-inspired intelligence debate - kzbin.info/www/bejne/ZqvbmmOuiKeCl7M - Pierre-Simon Laplace - Laplace's demon concept - en.wikipedia.org/wiki/Laplace%27s_demon - Tim Scarfe, Gary Marcus et al - GPT-3 access discussion - kzbin.info/www/bejne/n5TGlWtsq7Stadk - Gary Marcus - The Algebraic Mind book - mitpress.mit.edu/books/algebraic-mind - Noam Chomsky - MLST Noam Chomsky video - kzbin.info/www/bejne/l6nYeJmeabqFb7M - Gary Marcus - Neural networks and generalization work - www.sciencedirect.com/science/article/pii/S0010028598906946 - U.S. Congress - KOSA passage - www.congress.gov/bill/117th-congress/senate-bill/3663 - U.S. Congress - Section 230 - www.law.cornell.edu/uscode/text/47/230 - Gary Marcus - 'Is Generative AI Going to be a Dud?' article - garymarcus.substack.com/p/what-if-generative-ai-turned-out

[@federicoaschieri]

One critiques Silicon Valley and suddenly load of bots/ignorant fanboys appear in the comment section. Interesting. The view point of this channel has always been that LLMs are not advancing anywhere near to AGI. ChatGPT o1 does not solve at all the lack of a world model.

[@MatterandMind]

This may be Gary Marcus's biggest misconception. The World model is the foundation of how LLM works. This is confirmed by both theory and experiments. But many of the components of this model are far from perfect (partly due to the lack of multimodality).

[@federicoaschieri]

@@MatterandMind That's Marcus biggest insight. LLMs don't have world models, as the constant failure at common sense shows. Common sense, contrary to the misconception of many people, does not test reasoning, but the understanding of words, which LLMs don't have. Reality is much more complex than language, so an LLMs has only an axiomatic understanding of language, it does not and cannot understand its semantics.

[@ericbrown8855]

To all those who can’t stand a voice offering caution regarding a foundational shift in our collective human experience, I have to ask where the blind trust comes from in the folks building multi-million dollar bunkers and space rockets. Are they planning an exit strategy or just doing it for fun? Call me pedestrian, but I like it here. Messy? You bet. Room for improvement? Sure. But to simply cede our autonomy to a handful of aggressive, business-casual capitalists with a penchant for sci-fi books is hardly empowered or optimistic thinking.

[@GlennGaasland]

Most of his arguments seem to «push in open doors» in the sense of criticizing some basic misconceptions about LLMs, yet does not diminish the value of LLMs if this is understood. Like, it doesn’t have a “stable world model”. Well neither do I, nor would I want to, since no stable model could ever fit a dynamically shifting complex and strange world. All world models are at best extremely incomplete useful approximations. And why would an LLM need a world model similar to humans? It could have a very different approximate “world model” which could provide insight about patterns different from those humans perceive. Sure, LLMS are unreliable, as are we as humans. Again this is no major problem, since I don’t want the LLM to do all the thinking for me. I want it to complement and enhance and inspire my thinking, and there SHOULD be some weirdness and randomness to it. In specific cases where extreme accuracy is needed? Dont use LLMs! If you want perfect chess, use Stockfish. LLMs work best in a creative collaboration with humans where the humans make the major decisions, if you ask it dumb questions, you often get dumb answers, and if you ask profound questions you get profound answers. The real potential of LLMs are in enhancing human creativity and insight when working WITH humans.

[@tylermoore4429]

Gary says the same things over and over, and Tim did not push back on anything - I guess because he mostly agrees with Gary. What is a world-model? Do we have one? Do LLM's need it? I lean towards LLM's will not get us to AGI myself, but a deeper conversation about the philosophical/technical feasibility of the concept of AGI, about the strengths and limitations of LLM's and how far we can push them, and so on would have been more interesting than yet another repetition of Gary's LLM-bashing.

[@rossminet]

If you don't have a world model, you're probably not human. LLM is fill in the blanks on steroid, nothing more. "the brain is constantly generating and updating a "mental model" of the environment." Predictive coding, wikipedia.

[@benprytherchstats7702]

I'm pretty much on Gary's side, and I agree with most of what you've written here. I think your realistic view of LLMs is in conflict with what most of the tech CEOs are AI hypesters are selling. Sam Altman's out there talking about how AI is going to replace huge numbers of human jobs, OpenAI are waiving around o1's scores on tests designed for humans and trying to convince the public that it can "reason" as well as PhD physicists and mathmeticians, and most people seem to be eating it up. The mere fact that you're aware of LLMs' considerable limitations probably puts you nearer to Gary Marcus than to most tech journalists.

[@benprytherchstats7702]

@@tylermoore4429 FWIW they just did this two weeks ago: kzbin.info/www/bejne/pICZpHeha9uCZsk

[@GlennGaasland]

@@benprytherchstats7702 Yes and I would have loved hearing Gary talk about LMMs more from the creative and useful points of view, for example what about the best current LLMs (who can already do amazing things) IN COMBINATION with humans and other kinds of AI systems? Instead of saying obvious things we all have heard before and about everything thats misunderstood or wrong about these current systems…what if instead we focused on what was possible? Then I could see Gary have a lot to say and also be on a more inspiring path…because all likely futures from this moment must involve some mml system that we speak with continually about all kinds of things… The question «How could we use this technology WELL?» seems incredibly interesting and important now and far into the future…

[@egor.okhterov]

"People will see a single correct answer and assume that a system has correct underlying abstraction"

[@Hexanitrobenzene]

I don't understand such a strong criticism against Gary. His main points are true: 1. Current AI technology is fundamentally unreliable. Giving it more data won't solve that in principle. 2. Few large corporations are using AI hype to gain enormous influence. 3. AI is starting to cause societal harm and needs to be regulated.

[@richardnunziata3221]

what is interesting about GPT4 and chess is that given very simple rules on movement of pieces it still makes illegal movies as if it does not have a understanding of compositionality and systemecity but instead is just pattern prediction. Only with the ability to learn rules weather Pertrinets or other symbolic planning structures can it truly reason. Clearly if you asked the llm what are the rules of chess it would not have a problem to produce them so the disconnect is between knowledge and application of knowledge rules .

[@fburton8]

Tim, I love that you’re using an actual paper notebook rather than a phone or tablet. 😁

[@neurojitsu]

Love the clarity of Marcus' thought processes and explanations over such a broad range of aspects of AI's trajectory. On the morality discussion... I think the missing dimension to this discussion was the investment community as stakeholders and their typically "amoral" stance towards choosing their investment theses and all subsequent decision making. Maximising decisions for profit is always going to introduce moral dilemmas. As long as VCs, PE and funds are run by pepole who adopt an amoral stance motivated mainly by wealth accumulation (ie everyone in practice), there will be executives in AI companies who remain conflicted between their morals and their financial self interest. And as Taleb says, only financial independence can free us from this conflict - Gary being another case in point. In reality, entrusting the trajectory of AI to the morals of individual actors in the AI world is a fools game. Which is whey I think Gary is right that regulation is the only answer.

[@bujin5455]

Social media has definitely been a calamity, one I'm not certain we're going to recover from.

[@AsgerAlstrupPalm]

He talks like we all obviously get why we have to agree with him but I don't get it.

[@Charles-Darwin]

I agree, why are there so many people that take up the decision-making mantle. Without conveying to everyone what it is they see for pathways forward and what their research (in digestible terms, void of hard jargon) offers credence to their assertions, then most importantly, sampling the majority to confirm they can act...Instead, people clamber out of the shadows to insert themselves as the 'chief' and say their opinion is greater than all else. They "know better". How do we know he's not in the pocket of Intel or IBM (since he seems to argue to their side in the congressional hearings)? In those hearings, what struck me odd was the almost timidness of the IBM spokeswoman...she was incredibly reserved. Which indicates they didn't have near what Altman was bringing to that table to discuss. This guy was meant to be a 3rd party expert. Like what? 😆 They (anyone not Google or openai) have high af motivations to stall out progress of non-legacy companies so they could 'catch up' or churn butter with garbage derived products over years... Maybe even to keep the status quo of trickling updates to drag this all out (as in iPhone "revolutionizing" the phone every year) as long as possible. I wouldn't doubt it if they purchased people to be resistance factors. Not for sure, but the chicken soup this guy is serving is highly suspect. I think your Jedi sense is tuned correctly.

[@xemy1010]

So much of what Gary says here feels like he's taking down strawmen. "GPT-4 is not SOTA at playing chess" when literally no one claimed it was in the first place. "People are excited it can play chess at all". And? Just bizarre. He may as well release a new book titled "Bah Humbug" at this point. Also I find the double line of "AI pretty much sucks just as much as it did in 2021" and "I'm very disappointed that regulators aren't cracking down on AI" cynical and insincere.

[@oncedidactic]

@@xemy1010well the AI doesn’t have to be good for regulation to benefit us by curbing malignant corporate behavior

[@danielarista1352]

Well, he was a professor for 99.9% of his career.

[@priyakulkarni9583]

These triliannaire companies exploiting us! That is what he is telling. Don’t you get it? 😅 all he is saying is stop 🛑 the hype and take time to move ahead. 😅

[@BrianPeiris]

I appreciate this one. I've been vacillating on LLMs for a long while now and I dismissed Marcus several times in the past, but I see he's been right all along. I think part of it is that Marcus is so knowledgeable about this stuff that he takes his own knowledge for granted when giving these interviews so it just sounds like he's making blanket statements without backing them up. Viewers would have to go out of their way to find the full explanations in his blog posts (though he still might not explain his reasoning fully there either). Subbarao is better at walking laypeople through all the reasoning steps he's taken to reach his conclusions. Another problem is that there are equally compelling voices on the other side, who are usually more popular, so people are left trying to decide between polar opposite experts. It would really be great to see Marcus sit down with an LLM/AI proponent, Gawdat perhaps, and really walk them through his reasoning -- not in a debate or an adversarial manner. I want to see Marcus address the proponent's misconceptions one at a time, because I agree a lot of the hype is based on stacks of speculations about scaling laws, emergent behaviors, and Clever Hans projection effects.

[@MachineLearningStreetTalk]

Thank you!

[@BrianMosleyUK]

1:17:55 elegant way to say that Elon lacks integrity. I think that lots of us can see that, and it's disappointing.

[@boonkiathan]

Agree with many of his points, LLMs are impressive to humans and we are charmed by the emergent abilities, but is from being founded on any fundamentals but large bruteforce computations over data

[@MB-ue2rf]

I apologize but this line of criticism has the old dewy smell of symbolist indignation aghast at the connectionist dominance in AI, so that you hear such smatter about the fantastic stupidity of generative AI and that it will never lead to AGI, and yet it is to have the most far reaching consequences and should be rightly blamed for any exacerbation of the world's problems; problems that are hardly new and will always be with us.

[@astilen5647]

This was filmed before openai o1 model, and already this episode has not "aged well". Its hard to watch an expert coder devolve into opinionated philosophy, this is the age of recursive code after all, and we need our elders to keep up.

[@FamilyYoutubeTV-x6d]

I think he is not an expert coder either. I mean sure, he codes, but he is not an expert coder.

[@tomenglish9340]

Thanks. I really did not want to watch this if it were recorded prior to September 12, 2024 (which seemed rather like July 16, 1969, to this elder of yours, who not only keeps up, but revises his beliefs).

[@s1lli]

"Expert coder" Gary Marcus is definitely not that

[@astilen5647]

@@s1lli My point was experts should take accountability when they are wrong and make corrective appearances/actions. Will you take accountability when you google his name and find out how many AI dev teams he has directed? He is an expert coder.

[@astilen5647]

@@tomenglish9340 You can agree with people who are wrong, experts can be wrong, its nothing new. He is pro AI but anti ai hype, while at the same time signing ai regulation letters due to ai catastrophic potential? Those are conflicting notions, hype or no big deal???? This was very forgettable podcast, you wont have any reason to revisit it for knowledge and wisdom next year and you know this, I was only pointing out that it was not even worthy of rewatching it next Tuesday.

[@rickyale1618]

We can't trust big companies, and we shouldn't trust anyone until trust is earned. We should know that by now

[@heterotic]

It's good he realizes in retrospect that his actions in DC were naive.

[@MatterandMind]

What do you mean?

[@heterotic]

@@MatterandMind people should learn from their mistakes.

[@faster-than-light-memes]

Strange negative comment section. Thanks to dr. Marcus and this KZbin channel for staying on the ground of the facts

[@TheReferrer72]

What facts?

[@Matt-yp7io]

not strange, completely valid

[@benprytherchstats7702]

A lot of people really want to believe that LLMs are thinking and reasoning and acquiring "human-like" and "human-level" intelligence, so much so that they'll swallow silliness like "GPT-o1 did better than PhD physicists on a physics test OMG it's already smarter than us!" Gary's pointing out that this is an illusion, and most of the sorts of people who watch AI videos on KZbin don't wanna hear it. Notice that very few of the critical comments say anything specific. Just "LOL this guy sucks"

[@MiaNoble-w6c]

Thank you very much for a very commonsensical conversation about this new resource/tool!

[@johndewey7243]

This man can be wrong at the same time we are moving too fast. Don't let the message be killed by the messenger.

[@kowboy702]

I follow this page cause you give realistic outlooks of what AI is. Gary is awesome and I am quite disappointed in the comments.

[@itzhexen0]

The moral decline was about the time facebook came into existence. He's a little late.

[@Rockyzach88]

The start of "Moral decline" can be placed in many time periods. For instance I could say that it started in the late 70s upon the instantiating of "free market" neo-liberal capitalism and with it the xfer of power from middle class Americans to corporations. Additionally, commodification of basically everything in our lives has made it okay in the eyes of capitalists and the people riding their coattails to do away with ethics and "morality". To sum it up, big time media professionals in general have created a buffer between these exploiters and the population because they are playing the same game and their existence relies on it in the way it exists right now.

[@deep_AI_0]

everything he said made perfect sense to me. I dont understand why people are getting so pissed. GPT is in fact shit and orders of magnitude dumber than me at accomplishing something

[@mattwesney]

your naivety says otherwise, in fact its screaming... deafening, even.

[@deep_AI_0]

@@mattwesney whats your best argument on this topic

[@chronicallychill9979]

OpenAI's got a shit product and they deliberately make it shittier thru efforts to reduce the cost of running their poorly conceived heap of garbage. That you would draw your conclusions from your use of that one (again, shit tier) product is not a good sign for the validity of your opinion.

[@BrianMosleyUK]

Fascinating interview. Much to question and discuss on his opinions. I think that the hardware being built from the current hype will be needed to run the AI models which supersede generative AI. The capability is coming, and I don't think anyone inside Google, Anthropic, OpenAI would argue that they don't need further algorithmic breakthroughs. He brings the necessary focus to society, government and the rising oligarchies. I feel the same way about my new book "Uniting Hearts, Igniting Change : The Journey of Connection" - the world needs the message, just going to take some work to trigger the cascade effect of getting it read and recommended.

[@ArchonExMachina]

I think he extrapolates the particular weaknesses of current image models to all deep learning models. He is definitely an LLM hater and a symbolic AI guy. He doesn't understand deep learning with hidden layers at a deep level. It probably creates the ultimate database structure for connected data, that you can't have in an old-fashioned way. Also, with super-deep learning from an insane amount of data, I think there can be all sorts of representation built inside these models that we can't even fathom or appreciate yet. World modeling will be peanuts. He is only right in implying that a static model will not be enough for AGI, as agency and state management is required.

[@young9534]

the new o1 models are proof that LLM capabilities have definitely not peaked. Pretty sure this podcast was filmed before o1 release

[@wwkk4964]

I'm sure Gary will tell you how AI systems were regularly solving high energy astrophysics numerical problems in the 90s.

[@young9534]

@@wwkk4964 I give it 1-2 years before he has to start admitting he was wrong. But most likely he will just double down on the whole "LLMs are dangerous to society" thing

[@CharlesVanNoland]

I'll be impressed when an LLM can catch a ball, wash my dishes, or build a house.

[@wwkk4964]

@@CharlesVanNoland Is that something you heard from Stephen Hawking's ex?

[@steve_jabz]

His 2 hour lecture at the AGI summit just before o1 release is even funnier to watch in retrospect. Claims 4o and 4-turbo were supposed to be gpt-5 and asserts there's no way to scale reasoning beyond v4. As evidence he provides vibes and hand drawn graphs that go off the chart

[@richardnunziata3221]

why do we need representative democracy when we have a interconnected world people can vote or set vote preferences even have personal AIs that can vote based on its modeling of its user.

[@mahdipourmirzaei1048]

I understand his points on AI regulation. However, I still can't understand his specific concerns regarding the current capabilities of generative AI. In scientific debates, it's often valuable to maintain a degree of uncertainty and be open to very controversial topics like AGI. I believe that even with the current state of LLMs, without significant improvements, we might be able to automate a lot of things in our economy. So, he needs to respect this powerful technology.

[@Mynestrone]

Comment section seem to prefer oligarchs and libertarian kings over any type of democratic oversight. I know somebody will respond to this with "oh you think government" before you bother: 00:15:00

[@markmonfort29]

40 minutes in and what's the point. He's clever enough to understand there are nuances to these models or that what a model can do depends on the training... And he should be smart enough to know that combining models with other systems and techniques and even having humans in the loop can be good things and that we shouldn't expect the models to be all singing and all dancing... Is it complaints for the sake of complaints and to remain in the news cycle? I didn't like his views at first and then saw potential for him just being pragmatic with his take on AI and now we get this complaint fest.. I agree with caution in some ways but there's a way you can do that and not hold the space back by being so negative and obfuscative of the nuances. People will listen to him and agree which is the biggest harm he does I believe.

[@MarkDStrachan]

Yes, it was Hypernormalization.

[@BryanBortz]

Everyone should watch that.

[@jeanklein2198]

Comparing gen AI to Saturday Night Fever was a funny bit

[@LatentSpaceD]

Gotta pass on this one:) appreciate the show and the continued work:)

[@LearnWithBahman]

Can I have a link to his keynote?

[@MachineLearningStreetTalk]

kzbin.info/www/bejne/b2K2fGxmiMabfsU

[@chronicallychill9979]

LLMs aren't commercially relevant??? Is he trolling? I'm not in this space for a hobby, I got bills that have to be paid. And customers happy to pay them for me. Literally what is he saying?

[@alexandermoody1946]

The real phase that surveillance capitalism will see a remarkable shift towards is withdrawal as every aspect of our lives become more and more surveilled and relied upon for machine learning. An inflection position will become ever more concerning as the technology improves but human willingness to participate reduces. A combined purpose will be required where compensation for leveraged human interaction and digital data production becomes the natural progression where licensing and compensation is the motive factor for fairness in exchange for loss of privacy or instead of compensation the right to a private life which is seen as a fundamental human right. I do think that each of us will be marked perhaps comparable to a biblical sense of use but more over as a mechanism after that is seen in conjunction with human requirements rather than a point or position of enslavement. If not the erosion of human values and potential machine values will be vastly limited and that will not bode well for any people or machines.

[@dmitryabulkhanov8580]

Comparing chatbots like ChatGPT to airplane safety in one sentence is complete nonsense TBH

[@chastetree]

He's talking out of both sides of his mouth. He argues on the one hand that current LLMs suck and don't have any practical applications. On the other, he wants them regulated like airplanes or power tools, which have specific functional requirements that are amenable to quality control and regulation.

[@Irresistance]

@@chastetree Yes. Both attitudes can be true. Like, say......... you'd want generative AI to be stopped from creating assorted fake news, lies, celebrity pr0n, etc... yes? Not good. Very bad if advanced LLM's would be very good at that, *surely* ? Yet on the other hand, they are very much NOT thinking, agentic and sentient. Their output merely sometimes "fools" humans into kind of presuming/feeling as if they are. But that's an illusion. Also, LLM's haven't brought us closer to creating such agentic machines.

[@philmcgroin]

"It's seen god knows how many ants and it still doesn't understand" 😅

[@richardnunziata3221]

copyright laws are good as long as you apply them on generation not of training

[@johnkintree763]

I want to unite with people around the world in realizing the principles of the Universal Declaration of Human Rights and the Earth Charter. That's the first thing I plan on saying to a conversational global digital platform that can merge the knowledge and sentiment expressed in conversations with people into a shared graph and vector representation, for a form of collective intelligence. That's our moral ground.

[@psi4j]

Gary Marcus is a G

[@richardnunziata3221]

generative ai will give telcoms a boost when packaged but not as a subscription.

[@yikesawjeez]

while gary makes the occasional salient point, he's woefully underinformed about how the technology works or what the research thus far has indicated. We're nowhere close to hitting scaling law limits, llms (iirc even non-multimodal ones) likely do have and use world models, arguably convergent ones, almost none of the neuroscientific analogies he likes to allide to are relevant to large model behavior, 'trustworthiness' is easily corrected with rag + prompt engineering + a lora if you get desparate, the list goes on. I agree that the hype isn't great but he seems to me like someone who trained a few models for a masters' 10 years ago and spent a few hours with chatgpt then wrote a book being contrary (which sucks, 'cause i wouldnt mind someone with good communication skills and a dislike for hype talking to regulators if they knew what they were talking about)

[@drhxa]

Gary Marcus was surprisingly reasonable in this one. Great interview! And that's coming from someone who laughs at the vast majority of the nonsense Gary posts on X. But looking at the risks we face as a society, there are major risks both ahead and behind us As for whether GenAI is useful or not, I think that ship has sailed already and most people in tech realize that. Pushing a narrative that is useless will lose you more and more credibility especially with these new LRMs coming out like o1-preview

[@gigabake]

Weird to see this many sourpusses in MLST upset that a reasonable man doesn't buy into the hype

[@FamilyYoutubeTV-x6d]

He is a cynic and a pessimist. It's not about hype. It's about FACTS. Waymo is driving everywhere. LLMs are helping researchers fight manuscripts faster. LLMs are helping poor people write legal notices or legally based replies better. These systems are helping people learn things faster, people without mentors or teachers or money to afford one, etc. Marcus is naive and o1 proves him wrong on the reasoning side of things. Hallucinations can be minimized further with further symbolic exploration and more free memory, caching, etc. He is right about some moral aspects but that's what it takes to be the first to AGI honestly.

[@Apjooz]

Demand for five computers in the world still?

[@dcreelman]

Gary Marcus is far more lovable in conversation then he is on X.

[@BryanBortz]

This is one of the major flaws of micro blogging overall.

[@CodexPermutatio]

I see Gary Marcus, I listen.

[@kinngrimm]

16:20 meanwhile that inaction of the US senate leads to more and more voices even here in germany calling for less regulation as you don't have any and which causes our company lobbyist to now influence our politicians after the EU regulation was put in full motion. Just think about that, you don't get your shit together and others want to emulate that so they would not have any competativ disadvantages. Though i am of the mind seeing what wasn't invested into this area over decades that even with the right regulation we here in germany just have not buisness competing in that area, regulation or not. I think this topic is mostly used as vehicle for the otherwise disliked neoliberal talking points, to give the emperor a new set of cloths, rename, reframe, obfuscate and manipulate is the game.

[@LearnWithBahman]

You can already see bots in comments after him. Imagine this for everyone in 5-10 years.

[@jwadaow]

It's amazing how criticism is now solely attributed to bots.

[@user-be1jx7ty7n]

Somewhat disappointed that you are giving this man the time of day.

[@Rockyzach88]

Okay user-be1jx7ty7n

[@billykotsos4642]

Womp womp

[@Matt-yp7io]

its important to hear from people you profoundly disagree with / think are downright stupid

[@Apjooz]

@Matt-yp7io But this channel agrees with Marcus. Where are the true other opinions.

[@Matt-yp7io]

@@Apjooz the hosts maybe, not the audience

[@iamruki]

gary is the farthest thing from an AI expert. this guy is a joke.

[@AveryAshtynRebecca]

kahma AI fixes this (AI Headshots). Taming Silicon Valley: AI Expert Interview

[@kamiabayani]

no this guy sucks, he does nothing all day and just tweets about stoping progress

[@Rockyzach88]

Zealot. Enjoy your corporate overlords. Oh wait, you obviously already do. E: I do disagree with his statement about it being useless. It is unreliable for real operations right now though. As a personal tool it's great. It's obvious that Silicon Valley thinks they own the country.

[@Rockyzach88]

Lol my comment got deleted because I called this guy a zealot? Is zealot a bad word now?

[@Rockyzach88]

Gonna be fun watching -zealo...- I mean people like you learn that tunnel visioning is bad and counterintuitive.

[@egor.okhterov]

​@Rockyzach88 it's just youtube is malfunctioning. They have millions of comments incoming every second. Some of them get lost.

[@CalebAyrania]

Wait what is this Gary used decimate in the traditional sense...

[@schumiiit04]

Prof Gary Marcus is honest and is not trying to fool most people around AI Hype. I feel sorry for so many young talents who will feel cheated in the next 5 years, set aside for massive economic resources wasted on useless stuff/

[@l.halawani]

I clicked for the hat.

[@rey82rey82]

Enter: Strawberry

[@alarlol]

interesting how SO many people , from my viewpoint, are intelligent and bright and wise, even, in an area or some areas or almost all areas, sometimes. and then they start talking about politics :D

[@timealchemist7508]

I seriously made it through 6 min of the conversation and I’m so pissed off at this narcissist that I can’t listen further. Gary is so disingenuous and self serving he cannot grasp how wrong he is. Worst guest yet on this channel bar none.

[@BryanBortz]

Good stuff. Have him back soon.

[@MiaNoble-w6c]

AI = generative linguistic prediction engine with a psychopath God complex embedded into a thinking machine.

[@renanmonteirobarbosa8129]

stop paying taxes, the government will start listening real fast. Boycot the government LOL

[@jlljjl]

TLDR.... "elitist eating elitist" NOFCs

[@timealchemist7508]

This guy is going to ride the grift the whole way. Literally incapable of introspection

[@davidsvideos195]

I couldn't disagree with this guy more. What a naïve leftist.

[@F255123]

He really wrote the book because he's a grifter who wants money.

[@CodexPermutatio]

I can imagine you buying the book just so you can give it a proper critique.

[@chastetree]

I think with him is more about the attention.

[@furtsmagee1513]

This dude Gary just peddles speculative nonsense

[@derekcarday]

Gary Marcus is an absolute clown.

[@wwkk4964]

Even 2x speed is too slow for this one.

[@CarpenterBrother]

Skipped around a bit and it seems like he's repeating the same exact boring talking points. I don't have time for authoritarians, go next.

[@kowboy702]

Silicon Valley is full of Authoritarians bro

[@Rockyzach88]

Something tells me you probably are lol. It just has to be your particular flavor and it's probably a lot worse than "regulating corporations".

[@CarpenterBrother]

@@Rockyzach88 Nice attempt at writing a coherent sentence. You failed.

[@Rockyzach88]

@@CarpenterBrother Oh good thing, because I don't think you'd be able to handle it.

[@CarpenterBrother]

@@Rockyzach88 Yup, I've already lost couple of brain cells reading your previous remark. I wouldn't be able to handle more.

[@matthew1791]

He belongs in Europe

[@Rockyzach88]

And you belong in Russia.

[@fburton8]

Are you implying that he is un-American?

[@abstract9580]

No, theres a traditional split between people focussing on connectionism and GOFAI. Most people in the states have traditionally focussed on sub-symbolic/deep learning as an approach to AI, whilst their European research counterparts have traditionally focussed on symbolic systems. Comment section is generally sad place and misses the purpose of the show/ scientific discourse.

[@loopuleasa]

this guy is a professional internet troll he is not interested in the truth don't give him attention

[@shawnvandever3917]

Here is the problem with this self proclaimed "leading expert." How much has he worked with transformers? He hasn't! If we are talking rule based novelty vacuum cleaners then he can make some points. All he does is look for issues that thousands of people are already working on. He is such an expert but all he does is jack his jaws, he is not one of the thousands working on the problems. He is just a loud mouth "pick me."

[@JackHacksman]

IP lord masquerading as a consumer advocate.

[@Rockyzach88]

In this particular case (can't speak for his career as a whole as I don't really know who he is), he appears to be an advocate for people who don't like corporations having such free reign with technologies like AI without some safeguards and considerations of the downsides. Those aren't "consumers", those are citizens. They have lives outside buying things believe it not.

[@brendawilliams8062]

He’s calming the law makers down. While openly telling you. There’s no data replacement for your personal data. Assuming and assuring you it’s for training purposes only. If you can sue they are humble in a sense. If you can’t ok

[@primedeviation]

If Gary Marcus has his way Gary Marcus will have his way.

[@WillyB-s8k]

um, excuse me 'gary marcus', but my replika says we can all just inhale 600g of galaxy gas and everything will be simply wondrous. are you calling lir a liar? hmm? hmm? hmm? hmm? hmm? 😬

[@oncedidactic]

8:50 is the point, and it’s a good one.

[@rey82rey82]

6:30