François Chollet: Limits of Deep Learning | AI Podcast Clips

(more links below)

Podcast full episodes playlist:

Podcasts clips playlist:

Podcast website:

Podcast on iTunes:

Podcast on Spotify:

Podcast RSS:

Note: I select clips with insights from these much longer conversation with the hope of helping make these ideas more accessible and discoverable. Ultimately, this podcast is a small side hobby for me with the goal of sharing and discussing ideas. For now, I post a few clips every Tue & Fri. I did a poll and 92% of people either liked or loved the posting of daily clips, 2% were indifferent, and 6% hated it, some suggesting that I post them on a separate YouTube channel. I hear the 6% and partially agree, so am torn about the whole thing. I tried creating a separate clips channel but the YouTube algorithm makes it very difficult for that channel to grow unless the main channel is already very popular. So for a little while, I'll keep posting clips on the main channel. I ask for your patience and to see these clips as supporting the dissemination of knowledge contained in nuanced discussion. If you enjoy it, consider subscribing, sharing, and commenting.

François Chollet is the creator of Keras, which is an open source deep learning library that is designed to enable fast, user-friendly experimentation with deep neural networks. It serves as an interface to several deep learning libraries, most popular of which is TensorFlow, and it was integrated into TensorFlow main codebase a while back. Aside from creating an exceptionally useful and popular library, François is also a world-class AI researcher and software engineer at Google, and is definitely an outspoken, if not controversial, personality in the AI world, especially in the realm of ideas around the future of artificial intelligence.

Subscribe to this YouTube channel or connect on:

[lexfridman]

(more links below)

Podcast full episodes playlist:

Podcasts clips playlist:

Podcast website:

Podcast on iTunes:

Podcast on Spotify:

Podcast RSS:

[DasMahamud]

One for the Algo. Really enjoyed the conversation.

[SergAI]

Wait a minute, wasn’t a symbolic AI what Richard and Pied Piper created at the end of Silicon Valley???

[Filaxsan]

Great take on the illusion of intelligence vs real intelligence.

[HenkieIsNietGek]

Computers are optimizers. They are in their essence interpolators.

[IvanGarcia-cxjm]

I am not sure if this is a good rule of thumb to tell which applications could be done by deep learning or not. If the task can be performed by a human unconsciously after a lot of repetition, then it is deep learning friendly. If the human can't do it unconsciously and has to pay attention and reason about it, then deep learning might fail and some algorithms are required (along with deep learning helping).

[The_Conspiracy_Analyst]

There are problems where the growth of the size of whatever algorithm you are using relative to certain parameters in your problem set is critical. I know of some where NN absolutely WILL NOT WORK, and most of the researchers out there do not realize this yet. For example you can use a NN to solve some problem, but you are only solving for a finite set of the problem set, and you can't find analytic closure. So NN is useless there. Maybe ok to make some toy programs, that's it.

[DaveWhoa]

even if we manage to understand the human brain, and develop our own computer version of a human brain ... dont we still need to put it through 12 years of school?

[lasredchris]

Knowledge - recognize patterns in data
Neural nets don't generalize well
Turing test - tricking humans. Human perception

[johnalley]

Adding this guy to my NFoS list

(Not Full of Shit)

[kevinpurnell]

All current research into AI suffers from these problems. We should understand how human creativity works first then programming will likely come with ease

[jotatsu]

One question might be if we have logic circuit analogs in the brain, or it's just neural networks down to the bottom, arranged in some particular architecture. If we don't, then how can we develop logic and symbol manipulation but a neural network should not.

[The_Conspiracy_Analyst]

6:25 ChatGPT will never solve the turing test now. All you have to do is steer the converstaion to politically incorrect topics and gauge it's obviously canned responses. They AREN'T gonna change that. Therefore it can NEVER pass my turing tests. Just the way it is.

[InturnetHaetMachine]

Hey, can you do more insightful interviews like you did with Siraj Raval, ala Machine Learning rap and music videos. Clown.