Skip to main content
Gary Marcus

Gary Marcus

American psychologist and cognitive scientist

Nov 18, 2026

11:27
GitHub constantly
11:28
themselves.
11:28
They used to copy it from Stack Overflow and paste it and make a few edits.
11:34
And now the system will basically do that for them.
11:36
And so the kind of drudgery of it is replaced, and it's really genuinely useful for coders.
11:43
It has problems, too.
11:44
Sometimes it pastes in the wrong thing.

8 MINS LATER

20:12
briefly.
4:11
Where does then the company get the valuation from if it's burning through cash at the rate that you're suggesting it is?
4:17
Well, the biggest argument in favor is that their revenue has rocketed.
4:22
So they're going to make like 50 times what they made in 2025 and 2026.
4:28
But that's revenue, not profits.
4:30
So they're still not making profits.
4:32
Revenue went up and up and up really until about May when – There was this obsession in the field with something called token maxing.
4:41
And the idea was that you would reward your employees for using AI as much as possible.
8:26
So what sort of test is this for this technology?
49:20
this, is it the LLM that attaches to that? That's because previously- So the rough way to
49:24
think about it is- like a generate and filter.
49:28
The LLM generates a bunch of possibilities, and then the external reasoning system, which is lean, at least as far as I understand what's going on, goes through it and decides which is right.
49:39
That is the whole essence of being a neurosymbolic system.
49:42
Neural network generates, and there are different ways to do neurosymbolic, but this is a completely neurosymbolic solution.
49:47
Neural network generates something, the LLM generates something, and then something else says, does this make sense? So you're doing external verification with something that is not an LLM and is, you know, it's a version of a symbolic AI idea that goes back to the 1950s, right? Newell and Simon had general problem solver in the 1950s.
50:08
We should, by the way, actually solve some novel theorems.

6 MINS LATER

56:14
What's your, I just want to end here, like what is your AGI timeline? Because this is, honestly, Gary, it's the most hopeful I've heard you about AI since we've started speaking.
10:00
Do you want to share that with us?
10:02
Well, I mean, we don't know exactly what happened, but they added new software to old.
10:08
That new software almost certainly revolved around AI, and something went wrong when they added the new software to the old.
10:17
I had famously written a tweet.
10:19
Well, it wasn't famous until yesterday.
10:21
I had written a tweet four and a half years ago that said something like, it's going to be wild when they hook up generative AI to the air traffic control system.
10:34
And they never did that until Thursday, four and a half years later, and chaos immediately ensued, as I predicted.
10:58
So what exactly was this new AI thing?
7:42
The floor is yours to make your case.
7:44
Hello, Baltimore.
7:49
I grew up here, as many of you know.
7:51
My late mother lived here for 60 years, just a few blocks away, and I wish she could be here tonight.
7:57
Three years ago, I had a debate with Open Debate and with John, the parallels tonight about whether or not generative AI would on net benefit society.
8:05
The jury is still out, but between the massive wave of cyber hacks, deep fakes, and wholesale AI-generated disinformation, I have my doubts.
8:13
which is not to say that no AI could ever help society.

30 MINS LATER

38:36
I don't want to use the word ban because it's hard to control, but fundamentally having a rule that students can't use AI for some period of time, for a period of time during which they are developing the critical thinking that Cathy is talking about.
32:18
Now that he is gone, what are you predicting? What happens here?
32:23
It's almost impossible to predict.
32:25
I mean, chaos is the most obvious thing to predict.
32:27
So Jensen was a big risk taker, and he was putting money into all of these companies.
32:33
He's been replaced by a CFO who's more conservative, who might be like, why are we making all these bets when people are losing some confidence in the AI industry? They'll probably lose more.
32:42
confidence given that Jensen has exited the scene.
32:45
And so the circular financing thing could collapse and it could collapse very quickly.
Tim Seymour
Tim SeymourPANELIST
37:33
Yeah, that's a great
Steve GrassoPANELIST
35:57
Do you see any problem there?
36:00
I mean, I think the neoclouds are the most vulnerable here, right? They need the financing.
36:05
The financing is getting more expensive.
36:07
They're in trouble.
36:07
But I think one of the things we're wondering about, worrying about, is whether there's a contagion effect, right? If the neoclouds go down, does that drive NVIDIA down? Where else is the consequence? The other thing that came out this week is Bank of England had yet another one of these studies.
36:22
I think they looked at like 10,000, quote, decision makers showing, again, not that much productivity.
36:30
Having a $2 trillion build-out or whatever if you can't show the results is a problem.
36:55
Will they take NVIDIA down? What is the most concerning sort of transmission mechanism in your view? Is it that credit costs will go high and will raise costs for everybody across the board? Is it the circular financing aspect? Which part of it is the worst? It's
22:44
are you as scared as some people are about the potential for generative AI doing something truly off-piste, something truly rogue that would threaten large numbers of people, humanity? I am not worried about extinction.
23:02
A lot of people have talked about, oh, there's a 10% chance that AI might extinguish humanity.
23:06
That's a made-up number.
23:08
If AI started killing large numbers of people, people would fight back.
23:11
They would destroy data centers.
23:13
They would cut power lines.
23:15
If the robots come for you, the first thing you do is lock the door because robots can't even unlock doors.
24:48
It's finding itself caught out quite often, isn't it, by hallucinations? And what would your advice be to those businesses?
7:35
And yet you think that that's unlikely to happen because it's in nobody's interest.
7:41
There are a few different things that might happen.
7:43
And let me interject one other thing, which is slowdown is only one option.
7:47
Here's another option that I'm fond of or two of them.
7:50
One is I think we should actually pause the company OpenAI.
7:54
They should be put into receivership.
7:55
They have shown time and again that they are dishonest and incompetent.

18 MINS LATER

25:57
If AI is better at math than we are, should we let it be?
20:11
They hear powerful, they don't hear die.
20:13
Yeah, they don't hear die.
20:14
And I mean, maybe some of them are cynical.
20:15
They're like, "We're not really gonna die," and some of them are like, "It's...
20:18
But you know, I'm gonna make a lot of money before we do." I mean, different investors, you know, clearly think about it in different ways.
20:23
But they take that as a signal f- that an external party, which is not entirely external when you look at how all the money goes around, but anyway, the apparently external party is kind of validating how great the tech is, right? And I mean, I think though the, though the tech has improved in some ways, it's not as great as these people let on.
20:43
But it suits the doomers to have a narrative about how amazing the technology is.

14 MINS LATER

34:27
Meta's Muse, would this, would that be in this category?
2:16
So I want to get your thoughts on that warning, but more importantly, what worries you the most?
2:22
So let's start with what doesn't worry me so much, because that's the only good news I've got for you today, or might be, which is I'm not worried about extinction.
2:32
I think these numbers, like 10%, are basically made up.
2:35
And they always undersell humanity.
2:39
If you think about it, the notion of AI extinguishing humans, I mean, obviously, the heat death of the universe or whatever will eventually extinguish us.
2:49
We're not immortal as a species.
2:51
If we're talking about could AI kill us in the next five years, I think the chance is zero or very close to zero.

11 MINS LATER

14:13
There's
38:49
It's the fault of the people empowering them, and now the president is saying, "You boys, the people behind these, you can be in charge of yourselves."
38:59
Yeah.
38:59
Well, I mean, I wouldn't say that the White House is actually handling it.
39:04
The, the White House is handing it off, right? The White House has no policy here, um, except for one secret policy that I think they're just using to put their friends in, in power and keep others out.
39:16
Um, they literally haven't revealed that piece of policy.
39:19
So what they are doing is saying, "Hey, you take care of this." You know, even industries like the airline industry learned very early on that it's actually good to have regulation, and there's this crazy idea in Silicon Valley that has transferred over, uh, to Washington right now that you don't need regulation for AI, and this is just an accident waiting to happen.
39:40
In fact, some accidents have already happened.
40:07
Do you buy anything they're going to say? What should their goals be? And I want you to, to, to, to contrast what their goals should be and what their goals are.
2:12
Is there a legitimate threat or is there something else going on beneath the surface here that we need to worry about? Gary, this is right in your wheelhouse, so I figured we'd lead off with you.
2:20
Yeah, and there's a lot of bullshit to unpack there.
2:22
So I'll start by saying rogue AI is very anthropomorphic and very misleading.
2:29
It's a way to dispose of responsibility.
2:33
It wasn't open AI, it's their agents.
2:36
It's their fucking software that is having these problems, right? So if I built a car and I made the brakes out of cardboard instead of ceramic or whatever one does nowadays, you know, the car goes out of control.
2:48
Do you call it a rogue car? Well, that's what's going on is they made really lousy brakes, right? There are lots of things we know how to do in cybersecurity and they mostly didn't do them.

20 MINS LATER

22:40
Gary, it seemed like you wanted to chime in on this subject.
20:31
Is this a kind of meta way of... task allocation?
20:37
I wish it was.
20:39
So, neurosymbolic AI is just finding ways to integrate the two.
20:44
But I would call what you're describing orchestration.
20:47
Think about an orchestra.
20:48
You have all of these different Potential contributors, right? So you have your string section and you have your percussion and so on.
20:58
And the conductors and maybe the composer are bringing them in at different times.

19 MINS LATER

39:54
I guess the counter argument to that is you do that, it doesn't matter because China is going full speed ahead on this stuff anyway.
2:34
What do you see differently?
2:36
So I don't think extinction is a real likely risk for humanity.
2:42
All of the scenarios, the so-called doom scenarios where AI just wipes us out like in The Terminator, all assume that we won't fight back, and we will.
2:51
They all kind of, uh, miss some details about how we are geographically diverse, genetically diverse.
2:57
They're just fantasies and not realistic.
2:59
So extinction's not likely, but cat- catastrophe might be likely.
3:03
So we already had an incident where an AI, uh, reported something wrong and almost led to a war, and we will see more of those because these systems make errors and hallucinations.
5:35
What power should the government have that it doesn't currently have?
1:44
What do you see differently?
1:46
So I don't think extinction is a real likely risk for humanity.
1:52
All of the scenarios, the so-called doom scenarios where AI just wipes us out like in the Terminator, all assume that we won't fight back and we will.
2:01
They all kind of miss some details about how we are geographically diverse, genetically diverse.
2:07
They're just fantasies and not realistic.
2:10
So extinction is not likely, but catastrophe might be likely.
2:14
So we already had an incident where an AI reported something wrong and almost led to a war.
4:45
What power should the government have that it doesn't currently have?
1:44
What do you see differently?
1:47
So I don't think extinction is a real likely risk for humanity.
1:52
[coughs] All of the scenarios, the so-called doom scenarios where AI just wipes us out like in The Terminator, all assume that we won't fight back, and we will.
2:01
They all kind of m- uh, miss some details about how we are geographically diverse, genetically diverse.
2:07
They're just fantasies and not realistic.
2:10
So extinction's not likely, but cat- catastrophe might be likely.
2:14
So we already had an incident where an AI, uh, reported something wrong and almost led to a war, and we will see more of those because these systems make errors and hallucinations.
4:45
What power should the government have that it doesn't currently have?
1:44
What do you see differently?
1:47
So I don't think extinction is a real likely risk for humanity.
1:52
All of the scenarios, the so-called doom scenarios, where AI just wipes us out like in The Terminator, all assume that we won't fight back, and we will.
2:01
They all kind of, m- uh, miss some details about how we are geographically diverse, genetically diverse.
2:07
They're just fantasies and not realistic.
2:10
So extinction's not likely, but cat- catastrophe might be likely.
2:14
So we already had an incident where an AI, uh, reported something wrong and almost led to a war, and we will see more of those because these systems make errors and hallucinations.
4:45
What power should the government have that it doesn't currently have?
1:44
What do you see differently?
1:46
So I don't think extinction is a real likely risk for humanity.
1:52
All of the scenarios, the so-called doom scenarios where AI just wipes us out like in the Terminator, all assume that we won't fight back and we will.
2:01
They all kind of miss some details about how we are geographically diverse, genetically diverse.
2:07
They're just fantasies and not realistic.
2:10
So extinction is not likely, but catastrophe might be likely.
2:14
So we already had an incident where an AI reported something wrong and almost led to a war.
4:45
What power should the government have that it doesn't currently have?
37:42
Are they when they go into these rooms with these leaders, when they go into rooms with each other, are they start with the leaders? Are they saying, please regulate us because they want to be regulated? Or are they telling us that they're telling them, please regulate us? And are they going to these private dinners and saying, don't regulate us?
38:01
Well, I don't know what they say at the private dinner at the White House, but what they really want is to be regulated as little as possible, number one, and number two, to have so-called regulatory capture so they set the rules and other people can't catch up with them.
38:16
I had a picture, a cartoon that I borrowed in my sub stack of Dario at the top of the ladder trying to cut the ladder down so that nobody could catch up with him.
38:25
So ideally nothing, or if there's going to be something, it should keep them in first place.
38:29
That's what they want.
40:35
It's like they have no authority themselves, no agency themselves to slow it down.
40:42
Yeah, the SNL clip really captured the kind of contradictory nature of the things that Amadai says and the industry as a whole says.
40:52
If they really wanted to stop, they could.
10:45
Explain that for a sec.
10:46
So a neurosymbolic system is a hybrid of neural networks, which are basically supposed to look a little bit like the brain.
10:52
It's really just marketing.
10:54
They're not that much but like the brain.
10:56
We can go into that if you want, but, but that's what most of the stuff around now.
11:00
Today is these neural networks.
11:01
They're kind of statistical approximators.

34 MINS LATER

45:21
Like, will you explain the two if you might?
Dan Nathan
Dan NathanPANELIST
37:16
What would that mean for the ecos- Because he's basically saying what you're saying.
37:19
I mean, it was surprising to hear it from him.
37:21
Ezra Klein interviewed him, you know, for his New York Times podcast.
37:25
And, you know, I think Jensen doesn't want any new laws, and realizes that we do have product regulatory regimes, right? Like, you know, we took the Ford Pinto off the market because its gas tank exploded.
37:36
This is sort of like that.
37:37
Like, there could be real harm coming from this.
37:39
So far, I don't think there's, uh, been any severe harm, but we see that these systems can, you know, hack infrastructure, which could lead to, for example, a power grid going down that could take down a hospital and so forth.
Guy Adami
Guy AdamiPANELIST
38:24
About 17 years ago, too big to fail made its way into our lexicon, right? Uh, to Dan's question, are these companies now exactly that given how important they are to the entire ecosystem of the AI trade?
6:42
And what would you have to see to change your mind on that?
6:47
Um, I would want to see, to change my mind on that, some real scientific discovery that is beyond just kind of brute-forcing like we're seeing in math.
6:56
And I would want to see it in open-ended domains, which is really what science is and AI- AI is.
7:01
So the one place where we're seeing something like that is with math.
7:05
But with math, you can use these theorem provers and things like that that are not actually part of the neural network, although that's a separate conversation, um, and they can formally verify certain things.
7:16
And the problem with most science is it isn't like that.
7:19
Most of the interesting things in science, you can't just stick it in an equation and, and kind of check your, your formal proof.
speaker_7NARRATOR
11:26
OpenAI says they are, the way they describe it is, responding accordingly.
10:56
So Gary, um, what is your level of concern today?
11:01
Uh, well, I guess there are many different things to be concerned about, so it's, it's hard to, um, answer that.
11:07
Uh, part o- One set of concerns is about the AI itself, and the other is about politicians not understanding the problems.
11:14
So I am not at all concerned about extinction.
11:17
I like that you called it fringe.
11:19
It's a very popular fringe movement right now.
11:21
I think it distracts from the real concerns that we should actually have.
15:05
I- if AI, um, develops into, you know, a, a, a super system that can put everybody at risk, either with grids or with financial stuff or with drone attacks, doesn't that, um, affect everybody equally? I mean, if, if it's a r- if we're at risk here in the US, then China is at risk as well, and because we are all at risk in the same way that we were all at risk, um, during the nuclear age, still are, but really at risk in the moment, does that force these, um, hostile nations, China, the US, others, to come together to form some consensus because it is ultimately in everyone's best interest, including the ruling parties?
10:28
So Gary, um, w- what is your level of concern today?
10:33
Uh, well, I guess there are many different things to be concerned about, so it's, it's hard to, um, answer that.
10:39
Uh, part of...
10:40
One set of concerns is about the AI itself, and the other is about politicians not understanding the problems.
10:45
So I am not at all concerned about extinction.
10:48
I like that you called it fringe.
10:50
It's a very popular fringe movement right now.
14:58
And because we are all at risk in the same way that we were all at risk, um, during the nuclear age, still are, but really at risk in the moment, does that force these, um, hostile nations, China, the US, others, to come together to form some consensus because it is ultimately in everyone's best interest, including the ruling parties?

Scroll for more appearances

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.