Debunking the great AI lie | Noam Chomsky, Gary Marcus, Jeremy Kahn

Autocomplete is based on the previous few words. But modern language models use attention to condition on words arbitrarily far back. This lets them handle discourse coherence and long-distance dependencies. Autocomplete definitely can't do that! These models also creatively recombine things they've seen in new ways. They don't just regurgitate training examples. I can give it weird prompts like "explain photosynthesis in Shakesperean English" and it will generate something new. Good luck getting autocomplete to do that!

artificial anti-intelligence

One comment that struck me as relevant was made by Gary Marcus in which he said that "young cognitive science students are drawn away from the cognitive science into the GPt3 world where thay can make a lot of money...." This is a statement that explains where our effort truly lies. It is allowing the false idea of GPt3 to infiltrate the world as a valid idea in other words one that has "passed" all of the scientific tests about validity. therefore I think we have got to try to del with the underlying morality of the Free Market system of government and look into the idea of market control especially market control of economic necessities like childhood education and life development among people.

These people will criticise AI all the way to the top. Except Chomsky, who won't be there when it arrives. With every new model, Marcus will point out what it yet can't do. We know that, thank you. Why not do something useful instead ?

Should have set up automatic dictation/captions for Noam since he couldn't hear what they were saying. Maybe coulda got AI to do it LOL.

Noam's argument is too strong.

Noam notices that ChatGPT can learn how to produce text in "impossible" languages (meaning: languages with features that no natural human languages have), like languages with linear word order across sentence transformations. And therefore, because the systems can learn these "impossible" languages, they tell us nothing about learning or intelligence.

One problem: human beings can also learn languages that depend on strict linear word order. Like mathematical notation. So does that therefore imply that studying human beings can tell us nothing about natural language?

In other words, Noam's argument is irreparably flawed because it proves too much.

1. Current AI is Engineering but not Science
2. Science: what is the world like?
Engineering : How can I make something useful?
3. Its not even a good Engineering contribution
4. Dangers:
a. we are following the wrong path in AI. We need to look more at Humans. There have been fads in the past, this is another.
b. Perpetuate past data; past prejudices. Will produce disinformation will be devastating. Threat to democracy.
c. waste of capital - 100B$ wasted on self driving tech which will not happen (maybe a different version in 20 years)
d. people can take bad advice from these tools and do self harm
5. GPT is just autocomplete on steroids.
6. Mapping meaning to sentence is needed. Not just being able to make a sentence.
7. Distinction is between Neural Networks vs. Knowledge based approach - we have drifter towards NN about 50 years ago. Therefore, we have lot of knowledge available but don't know how to put it into systems.
8. Understanding/meaning does not magically 'emerge' from data patterns. We have (innate) constructs in our mind that enable us to find meaning in data (from plato)
9. We get it right 75% of the time, and we think we are making progress. We are not. We saw it with driverless car tech.
10. Future steps should potentially involve incorporating innateness into AI. Maybe a mix of NN and Abstract knowledge.11.

This is nonsense, they don't even understand neural networks. Old and naive guys spouting off. Self driving cars are happening, AI is on an exponential curve, the future is AI , and hopefully AI and human hybrids.

You thought that was John Oliver, didn't you.

Noam Chomsky: An opinion about everything, knowledge about the same? Not so much.

The dude from Saw spitting knowledge

Chomsky: Language and thought are virtually indistinguishable.

That is not correct.

They've been proven wrong on almost every skeptical claim. And it didn't even take a year.

respect for both gentlemen, but I have a feeling like they are looking at 2 year old baby (GPT4) and asking why the baby is not understanding the world? Something is terribly wrong with the baby, we need to discard it and use our knowledge of NLP to hand-code a new intelligent baby from scratch (just like they tried for 40 years) because this new baby GPT4 clearly doesn't work.

The dinosaurs will die. The memories of those with useful contributions to language and humanity will only survive by the significance of contributions. Neither of these two academics has left anything of import that will transform society by an unexpected force and semantic distillation, synthesizer, and generator of rational intuitive gestalt discourse and recourse. The reductionissm of this video only seerves to diminish groundbreaking experiments as propaganda and only THEY are the true knowers and value givers to society of anything important and language related. Give us a break and quiet already please. Contribute something besides anarcho syndoclisn and other ideological tropes and branches of political biases rather than Einsteinian departure from personal hopes and desires.

Cannot stand dealing with it in ANY form used as voice replacement. Its always SLOW, missspeaks with poor pronunciation, needs simple words repeated. Then still gets it wrong. Business men are fools deleting all forms of personal customer service. Gonna come back at you. Notice that 1,000,000,000 workers all REFUSE to return to offices.

Gary Marcus has been saying AI can't do this and it can't do that forever, and keeps shifting the goalposts whenever it hits certain milestones.

But surely this is why we are not calling it A.G.I. yet ?

ai has helped me: generate over 1000 images about specific topics programatically, sped up productivity many times on my various programming projects, helped me diagnose medical issues, helped me understand parts of the legal system that I needed to know about, oh and im sure there's quite a few things i can't remember right now. you can say you don't like ai but don't say it's a lie. Certainly don't say it's a lie while also saying it's going to end the world. Your arguments are not coherent.

well, this didn’t age well as others also mentioned it lmao

Some of the comments Gary and Noam made don't seem to be aging well (most notably on self-driving cars). We'll see in the next 1-2 years. That said, it's so important to have these different perspectives as it can help us develop better systems.

Without solving the so-called Hard Problem of Consciousness, not a single academic or scientist can say anything meaningful or consequential about "AI" beyond the "Event Horizon of the Technological Singularity".

Not only does said problem remain unresolved throughout the entire global academic and scientific institutions, the question has not even been properly or adequately structured.

If one does not know how to ask the right questions then the"Hard Problem of Consciousness" will remain unresolved and as such will stimy all attempts to develop the requisite intellectual, emotional and psychological instruments to successfully navigate the Event Horizon of said Singularity, the effect being, total extinction of the human species within a very short time.

AI is a Dead mantra,
Superstition and Illiteracy,
the Simple Nature of Intelligence,
can Never be artificial.
It is all about Programmed Consciousness.
A book is also programmed Consciousness,
Frozen Memory.
The Danger is, the Fooling of more than a billion School children,
this was Not possible in the Middle Age.

The example of the wrong order up and cubes is only a smal failure, the can be corrected with just one single correction sample in a "memory" model. Until now it was not allowed to overrul the system but this will change! Chatgpt is not ruled base system but it can become one and you make the rules and later you can teach this system to find the rules by it self and how thinks are connected to gether.

Not sure what's more dumb, the megalomaniac techbros with main-char-disorder, or the people in the video 😆

It's too obvious that Gary is not in the same league with Noam.

This video actually aged well. The reason gpt 4 is better than gpt 3 is because they used more data. And if they used all the data they could get, where are they gonna get new data for gpt 5 and will it significantly improve?

AI is stupid. I make some test on GPT and is so stupid, instead give a proper answer it give you a ton on sample and point.
i don't want read all this crap i want get an answer. 100% true.

This aged horribly... esp Marcus. What Noam said might have some valid philosophical value, but now shown to be too pessimistic and dismissive.

All the criticism ignores the fact that these LLMs are turning out to be immensely useful and boost our productivity like no other technology before them.

Anybody who denies that simply doesn't know how to use LLMs properly which is a real puzzle, because these are the easiest, most straightforward interfaces to exploiting computing power we have yet to come up with.

I came here for John Oliver!

Show me an ai that can produce anything with zero errors

@dallassegno AI is a wide range of things. So yes and no.

These video thumbnails swing back and forth. There’s a definite pattern. Two weeks of the end of the world, then two weeks of it’s all hype .

I've been saying for some time that the real danger of AI is imagining it is something it isn't.

Incredible that Noam has retained his sharpness even to this day.

It is ironic how the live transcript on this clip consistently uses "gnome" instead of Noam...

2024: Tesla FSD is changed.

A machine will never help a mother know when to let her son cry or when to help him develop self-control. Information is not wisdom.

Gary, gpt is not auto select on steroids jfc imagine calling a LLM a LSTM 😅

Those smart guys surely know something bad is going to happen assuming the amount of water tanks behind their backs they have prepared.

Aged like milk. "a red cube on top of a blue cube." Couldn't get it to draw the reverse. Technology moves fast.

AI passes the Turing test easily, so we move the goal posts! It's even surpassed the criticisms of this talk from a year ago, so formulate some new criticisms. You have to appreciate that the Turing test seemed like a likely impossibility for many decades

The least AI deals with is Language and how the machine chooses one word and not another. It is what is behind it, the Mathematical Functions are what are operating behind it, articulating the actions of the neural network. The main of these functions, said by Geoffrey Hington and Illya Sutskever, is the Cost Function, the network seeks the greatest profit at the lowest cost function. And this is proving to work for all modes of information, language, audio, video, movements, touch, to recognize smells and tastes. This has long since surpassed language, it has already passed through that station 🚉 and these people seem not to realize that neural networks are a chain reaction whose knowledge of the world is strengthened and grows with Multimodality, with Data, with the Growth of the synapses and the increase in the power of the GPUs (Flops). This shows no signs of stopping or stagnating at all as Raymond Kurzweil predicted. That's really what's happening, at least in this phase.

OMG how i DESPISE everything AI ....i despise what AI does to the whole concept amd proces of lifelong studying , and mastering - "BEING an ARTIST" musically or visually - now because of this satanic aproach , called AI - 5000 years of diverse culture is OUT the window (all pun int.) and we will NEVER again , know , how many % human endevour is behind an "Artist´ s" work ,- And nobody seems to givafuck,-

There are people who can't do art adopting the title of "artist" because they can use this stupidity to create cheese images. It's not art, folks. No feeling, no soul. It's the opposite of art, the opposite of beauty.

It’s not intelligence. It’s hyped up data engineering to con investors….😊

Mark my words.... we either created our saving grace , or our replacement... A.i is a beautiful thing , but once it becomes concuss we may be in trouble once A.I finds out how evil n disgusting humans truly are.

We need to experiment with A.I as in feed it everything n truly try to make it human. But keep it contained and see what it does n tries to do.... this way you'll kno if it's worth it or not