episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It's very conserved across species, something
https://karpathy.ai/lexicap/0013-large.html#00:53:53.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
that evolution discovered early.
https://karpathy.ai/lexicap/0013-large.html#00:53:56.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It may be the reason why babies tend
https://karpathy.ai/lexicap/0013-large.html#00:53:59.040
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
to look in the first few days to moving objects
https://karpathy.ai/lexicap/0013-large.html#00:54:01.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and not to not moving objects.
https://karpathy.ai/lexicap/0013-large.html#00:54:06.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Now, moving objects means, OK, they're attracted by motion.
https://karpathy.ai/lexicap/0013-large.html#00:54:08.320
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
But motion also means that motion
https://karpathy.ai/lexicap/0013-large.html#00:54:12.200
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
gives automatic segmentation from the background.
https://karpathy.ai/lexicap/0013-large.html#00:54:15.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So because of motion boundaries, either the object
https://karpathy.ai/lexicap/0013-large.html#00:54:20.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
is moving or the eye of the baby is tracking the moving object
https://karpathy.ai/lexicap/0013-large.html#00:54:25.360
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and the background is moving, right?
https://karpathy.ai/lexicap/0013-large.html#00:54:30.600
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Yeah, so just purely on the visual characteristics
https://karpathy.ai/lexicap/0013-large.html#00:54:32.800
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
of the scene, that seems to be the most useful.
https://karpathy.ai/lexicap/0013-large.html#00:54:36.040
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Right, so it's like looking at an object without background.
https://karpathy.ai/lexicap/0013-large.html#00:54:37.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It's ideal for learning the object.
https://karpathy.ai/lexicap/0013-large.html#00:54:43.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Otherwise, it's really difficult because you
https://karpathy.ai/lexicap/0013-large.html#00:54:45.760
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
have so much stuff.
https://karpathy.ai/lexicap/0013-large.html#00:54:48.760
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So suppose you do this at the beginning, first weeks.
https://karpathy.ai/lexicap/0013-large.html#00:54:50.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Then after that, you can recognize object.
https://karpathy.ai/lexicap/0013-large.html#00:54:55.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Now they are imprinted, the number one,
https://karpathy.ai/lexicap/0013-large.html#00:54:58.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
even in the background, even without motion.
https://karpathy.ai/lexicap/0013-large.html#00:55:02.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So that's, by the way, I just want
https://karpathy.ai/lexicap/0013-large.html#00:55:05.800
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
to ask on the object recognition problem.
https://karpathy.ai/lexicap/0013-large.html#00:55:08.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So there is this being responsive to movement
https://karpathy.ai/lexicap/0013-large.html#00:55:10.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and doing edge detection, essentially.
https://karpathy.ai/lexicap/0013-large.html#00:55:13.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
What's the gap between being effective at visually
https://karpathy.ai/lexicap/0013-large.html#00:55:16.760
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
recognizing stuff, detecting where it is,
https://karpathy.ai/lexicap/0013-large.html#00:55:21.600
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and understanding the scene?
https://karpathy.ai/lexicap/0013-large.html#00:55:24.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Is this a huge gap in many layers, or is it close?
https://karpathy.ai/lexicap/0013-large.html#00:55:27.640
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
No, I think that's a huge gap.
https://karpathy.ai/lexicap/0013-large.html#00:55:32.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I think present algorithm with all the success that we have
https://karpathy.ai/lexicap/0013-large.html#00:55:35.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and the fact that there are a lot of very useful,
https://karpathy.ai/lexicap/0013-large.html#00:55:42.040
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I think we are in a golden age for applications
https://karpathy.ai/lexicap/0013-large.html#00:55:45.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
of low level vision and low level speech recognition
https://karpathy.ai/lexicap/0013-large.html#00:55:48.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and so on, Alexa and so on.
https://karpathy.ai/lexicap/0013-large.html#00:55:53.720
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
There are many more things of similar level
https://karpathy.ai/lexicap/0013-large.html#00:55:56.800
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
to be done, including medical diagnosis and so on.
https://karpathy.ai/lexicap/0013-large.html#00:55:58.840
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
But we are far from what we call understanding
https://karpathy.ai/lexicap/0013-large.html#00:56:02.040
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
of a scene, of language, of actions, of people.
https://karpathy.ai/lexicap/0013-large.html#00:56:05.600
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
That is, despite the claims, that's, I think, very far.
https://karpathy.ai/lexicap/0013-large.html#00:56:11.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
We're a little bit off.
https://karpathy.ai/lexicap/0013-large.html#00:56:18.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So in popular culture and among many researchers,
https://karpathy.ai/lexicap/0013-large.html#00:56:19.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
some of which I've spoken with, the Stuart Russell
https://karpathy.ai/lexicap/0013-large.html#00:56:23.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and Elon Musk, in and out of the AI field,
https://karpathy.ai/lexicap/0013-large.html#00:56:25.640
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
there's a concern about the existential threat of AI.
https://karpathy.ai/lexicap/0013-large.html#00:56:30.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
And how do you think about this concern?
https://karpathy.ai/lexicap/0013-large.html#00:56:34.520
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
And is it valuable to think about large scale, long term,
https://karpathy.ai/lexicap/0013-large.html#00:56:40.000
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
unintended consequences of intelligent systems
https://karpathy.ai/lexicap/0013-large.html#00:56:45.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
we try to build?
https://karpathy.ai/lexicap/0013-large.html#00:56:50.360
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I always think it's better to worry first, early,
https://karpathy.ai/lexicap/0013-large.html#00:56:51.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
rather than late.
https://karpathy.ai/lexicap/0013-large.html#00:56:56.000
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So worry is good.
https://karpathy.ai/lexicap/0013-large.html#00:56:58.640
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Yeah.
https://karpathy.ai/lexicap/0013-large.html#00:56:59.640
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I'm not against worrying at all.
https://karpathy.ai/lexicap/0013-large.html#00:57:00.400
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Personally, I think that it will take a long time
https://karpathy.ai/lexicap/0013-large.html#00:57:03.000
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
before there is real reason to be worried.
https://karpathy.ai/lexicap/0013-large.html#00:57:09.520
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
But as I said, I think it's good to put in place
https://karpathy.ai/lexicap/0013-large.html#00:57:15.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and think about possible safety against.
https://karpathy.ai/lexicap/0013-large.html#00:57:19.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
What I find a bit misleading are things
https://karpathy.ai/lexicap/0013-large.html#00:57:24.360
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
like that have been said by people I know, like Elon Musk,
https://karpathy.ai/lexicap/0013-large.html#00:57:27.360
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and what is Bostrom in particular,
https://karpathy.ai/lexicap/0013-large.html#00:57:31.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and what is his first name?
https://karpathy.ai/lexicap/0013-large.html#00:57:35.240
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Nick Bostrom.
https://karpathy.ai/lexicap/0013-large.html#00:57:36.800
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Nick Bostrom, right.
https://karpathy.ai/lexicap/0013-large.html#00:57:37.400
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
And a couple of other people that, for instance, AI
https://karpathy.ai/lexicap/0013-large.html#00:57:40.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
is more dangerous than nuclear weapons.
https://karpathy.ai/lexicap/0013-large.html#00:57:44.080
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I think that's really wrong.
https://karpathy.ai/lexicap/0013-large.html#00:57:46.880
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
That can be misleading.
https://karpathy.ai/lexicap/0013-large.html#00:57:50.400
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Because in terms of priority, we should still
https://karpathy.ai/lexicap/0013-large.html#00:57:52.680
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
be more worried about nuclear weapons
https://karpathy.ai/lexicap/0013-large.html#00:57:56.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and what people are doing about it and so on than AI.
https://karpathy.ai/lexicap/0013-large.html#00:57:59.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
And you've spoken about Demis Hassabis
https://karpathy.ai/lexicap/0013-large.html#00:58:05.600
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
and yourself saying that you think
https://karpathy.ai/lexicap/0013-large.html#00:58:09.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
you'll be about 100 years out before we
https://karpathy.ai/lexicap/0013-large.html#00:58:12.840
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
have a general intelligence system that's
https://karpathy.ai/lexicap/0013-large.html#00:58:16.440
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
on par with a human being.
https://karpathy.ai/lexicap/0013-large.html#00:58:18.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Do you have any updates for those predictions?
https://karpathy.ai/lexicap/0013-large.html#00:58:20.600
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Well, I think he said.
https://karpathy.ai/lexicap/0013-large.html#00:58:22.520
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
He said 20, I think.
https://karpathy.ai/lexicap/0013-large.html#00:58:24.080
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
He said 20, right.
https://karpathy.ai/lexicap/0013-large.html#00:58:25.080
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
This was a couple of years ago.
https://karpathy.ai/lexicap/0013-large.html#00:58:26.200
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I have not asked him again.
https://karpathy.ai/lexicap/0013-large.html#00:58:27.680
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
So should I have?
https://karpathy.ai/lexicap/0013-large.html#00:58:29.160
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Your own prediction, what's your prediction
https://karpathy.ai/lexicap/0013-large.html#00:58:31.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
about when you'll be truly surprised?
https://karpathy.ai/lexicap/0013-large.html#00:58:36.000
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
And what's the confidence interval on that?
https://karpathy.ai/lexicap/0013-large.html#00:58:38.880
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It's so difficult to predict the future and even
https://karpathy.ai/lexicap/0013-large.html#00:58:43.000
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
the present sometimes.
https://karpathy.ai/lexicap/0013-large.html#00:58:45.760
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It's pretty hard to predict.
https://karpathy.ai/lexicap/0013-large.html#00:58:47.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
But I would be, as I said, this is completely,
https://karpathy.ai/lexicap/0013-large.html#00:58:48.480
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I would be more like Rod Brooks.
https://karpathy.ai/lexicap/0013-large.html#00:58:53.360
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
I think he's about 200 years.
https://karpathy.ai/lexicap/0013-large.html#00:58:56.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
200 years.
https://karpathy.ai/lexicap/0013-large.html#00:58:58.960
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
When we have this kind of AGI system,
https://karpathy.ai/lexicap/0013-large.html#00:59:01.560
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
artificial general intelligence system,
https://karpathy.ai/lexicap/0013-large.html#00:59:04.880
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
you're sitting in a room with her, him, it.
https://karpathy.ai/lexicap/0013-large.html#00:59:06.920
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Do you think the underlying design of such a system
https://karpathy.ai/lexicap/0013-large.html#00:59:12.840
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
is something we'll be able to understand?
https://karpathy.ai/lexicap/0013-large.html#00:59:17.120
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
It will be simple?
https://karpathy.ai/lexicap/0013-large.html#00:59:19.080
Tomaso Poggio: Brains, Minds, and Machines | Lex Fridman Podcast #13
Do you think it'll be explainable,
https://karpathy.ai/lexicap/0013-large.html#00:59:20.480