Yann LeCun is the Chief AI Scientist at Meta, professor at NYU, Turing Award winner, and one of the seminal researchers in the history of machine learning. Please support this podcast by checking out our sponsors: - Public Goods: https://publicgoods.com/lex and use code LEX to get $15 off - Indeed: https://indeed.com/lex to get $75 credit - ROKA: https://roka.com/ and use code LEX to get 20% off your first order - NetSuite: http://netsuite.com/lex to get free product tour - Magic Spoon: https://magicspoon.com/lex and use code LEX to get $5 off EPISODE LINKS: Yann's Twitter: https://twitter.com/ylecun Yann's Facebook: https://www.facebook.com/yann.lecun Yann's Website: http://yann.lecun.com/ Books and resources mentioned: Self-supervised learning (article): https://bit.ly/3Aau1DQ PODCAST INFO: Podcast website: https://lexfridman.com/podcast Apple Podcasts: https://apple.co/2lwqZIr Spotify: https://spoti.fi/2nEwCF8 RSS: https://lexfridman.com/feed/podcast/ YouTube Full Episodes: https://youtube.com/lexfridman YouTube Clips: https://youtube.com/lexclips SUPPORT & CONNECT: - Check out the sponsors above, it's the best way to support this podcast - Support on Patreon: https://www.patreon.com/lexfridman - Twitter: https://twitter.com/lexfridman - Instagram: https://www.instagram.com/lexfridman - LinkedIn: https://www.linkedin.com/in/lexfridman - Facebook: https://www.facebook.com/lexfridman - Medium: https://medium.com/@lexfridman OUTLINE: Here's the timestamps for the episode. On some podcast players you should be able to click the timestamp to jump to that time. (00:00) - Introduction (06:58) - Self-supervised learning (17:17) - Vision vs language (23:08) - Statistics (28:55) - Three challenges of machine learning (34:45) - Chess (42:47) - Animals and intelligence (52:31) - Data augmentation (1:13:51) - Multimodal learning (1:25:40) - Consciousness (1:30:25) - Intrinsic vs learned ideas (1:34:37) - Fear of death (1:42:29) - Artificial Intelligence (1:56:18) - Facebook AI Research (2:12:56) - NeurIPS (2:29:08) - Complexity (2:37:33) - Music (2:42:28) - Advice for young people
Yann LeCun, chief AI scientist at Meta and professor at NYU, discusses his pioneering work in AI and machine learning, as well as his thoughts on current AI research and its future implications.
A runner reviews essential and classy-looking running shoes that are suitable for both fast and slow pace runs, as well as with a suit.
This podcast discusses the challenges and importance of using reinforcement learning to solve problems that require understanding of intuitive physics. Reinforcement learning involves providing the machine with a single scalar for every trial.
The current approaches to self-supervised learning in NLP are oversimplified and do not adequately represent uncertainty, multiple outcomes, or a continuum of possibilities in the distribution of words, leading to a need for a more compressed representation.
Self-supervised learning tries to mimic how the human brain works in terms of prediction as an essential mechanism for task-independent learning, updating and adjusting models based on sensory information.
Gradient-based model predictive control can be used to help intelligent agents do planning, reasoning, and learning, allowing agents to interact with complicated physical systems and environments.
Humans use logical reasoning less often than expected but have the ability to build models of the world through self-supervised learning and reasoning from data, in addition to social interaction and language skills.
This podcast episode discusses the development of image recognition systems and their ability to be trained through self-supervised or weakly supervised methods. It also explores the challenges of filtering malicious signal in self-supervised learning.
Data augmentation is the practice of artificially increasing the size of a training set by distorting images without changing the nature of the image, which can improve the performance of neural networks in image recognition tasks.
The type of data augmentation needed for computer vision tasks should depend on the specific tasks and goals of the system. Currently, standard data augmentation techniques are only appropriate for object recognition or image classification, and cannot be used for tasks like object detection and localization.
The speaker proposes using generative data augmentation techniques to improve the process of visual similarity learning and suggests techniques such as interactive masking and corruptions to improve the training procedure.
This podcast discusses the various techniques used for computer vision, speech recognition and natural language understanding, including end-to-end deep learning systems, convolutional nets, and conventional computer vision techniques.
In order to train a causal model of the world that allows for the prediction of the consequences of actions, a system must interact with the world and learn from its mistakes. The idea of consciousness and its role in artificial intelligence parallels questions from centuries ago about the workings of the eye.
The things we believe to be hardwired in our brains may actually just be learned behaviors that are easy to pick up. This includes beliefs about religion and concepts like immortality that we may construct to ease our minds.
The discussion of whether intelligent machines and systems should have the same rights as human beings is still an ongoing debate. As humans continue to build more sophisticated AI systems, the development of an underlying theory of intelligence that encompasses not just artificial but also human and biological intelligence in general is crucial in defining the relationship between AI and human rights.
This episode delves into the ethical and legal implications of declaring robots as sentient beings, discussing the consequences of their mistreatment and the challenges in defining true intelligence in artificial systems.
The former director of Facebook's AI research division discusses his experience setting up the division and fostering a culture of open research. He also mentions the importance of creating urgent deadlines to achieve the seemingly impossible, and talks about his current work on self-supervised learning.
With Facebook changing its name to Meta, the question remains whether people will be interested in using virtual reality for extended periods of time, and whether the negative effects of social media can truly be attributed to it.
The speaker compares social media to the printing press, stating that both have advantages and disadvantages when it comes to spreading information and creating division.
The goal of training a system to predict video is to learn a representation of two video clips that hold as much information as possible, but exclude the elements that are hard to predict. This method involves predicting an abstract representation of pixels, instead of actual pixels themselves.
This podcast explores a new system for reviewing scientific papers in which papers are evaluated based on the worthiness of their ideas rather than just the quality of their experiments or protocols, and how relying on the opinions of trusted groups or individuals can save time.
The podcast explores modern physics, quantum physics and their ability to recover information lost in a black hole. It also touches on patterns in physical systems, chaotic systems, the emergence of life and complexity theories.
The speaker emphasizes the importance of learning fundamental sciences such as quantum mechanics and statistical physics over consumer technologies as they have a deeper impact on real-life applications such as drug design and machine learning.
In this podcast, Jan Lekun talks about the potential of machine learning to revolutionize industries while also speaking on the challenges and potential downsides of the technology.