Results 1 to 1 of 1

Thread: Lex Fridman x Ilya Sutskever, leading researcher in Deep Learning

  1. #1

    Lex Fridman x Ilya Sutskever, leading researcher in Deep Learning

    This is interview is pretty technical. Lex is an expert in AI and Sutskever is one of the leading personas in the field of AI and Deep Learning, worldwide. They don't pull any punches for the listener. So just be forewarned.

    That said, I think it is valuable for the wider public to get exposure to this discussion. Ignore the technicalities and listen to them discuss the practical implications of Deep Learning for the future. There is an opinion that many people hold that Deep Learning is just a buzzword. That this is yet again the same situation of the boy who cried wolf... "Skynet is here!"

    I am an engineer working in the high-tech field and I think it is important for the public to understand that this time really is different. There are no major AI tasks which Deep Learning has not either conquered or at least cracked open. That last point is impossible to over-emphasize. That Deep Learning is not yet having lucid conversations with humans or plotting nuclear world domination is irrelevant. It's the wrong measure of progress. Deep Learning has already revolutionized the technology that you carry with you in your pocket, in a thousand different ways.

    Perhaps the most noticeable change for end-users is the massive increase in reliability of the speech-to-text feature. Deep Learning is why your phone doesn't need to train to understand your voice or your accent and yet it is able to recognize what you say with accuracy that is on par with a human (with occasional glitches, of course). The change from legacy speech-to-text algorithms to Deep Learning-based speech-to-text is night and day. And it occurred without end-users being aware of how it happened.

    In practical terms, I think that the public should think of Deep Learning as a general-purpose "compute-accelerator". Deep Learning takes a given pool of compute resources and makes them able to solve problems that are factors of millions, billions, or more, harder than legacy systems could solve even with access to the same compute resources. That is the key. For decades, we have had the capacity to scale up the available compute hardware as much as the market will bear. But due to the slow process of writing software by hand, the market has always been capped by the capacity of software designers to utilize the available hardware. Deep Learning has changed all that. For any typical Deep Learning task, if you give me 1,000 times more compute, I will scale up your pipeline by nearly 1,000 times. That linear scaling is unprecedented in the history of computers. More available compute has always meant greater time-to-market while the software engineers code up algorithms to utilize the available compute resources. Not anymore.

    If you'd like to play with GPT-2, OpenAI's controversial text-prompt engine, click here:

    This engine was considered so effective at generating artificial, human-like English, that OpenAI decided not to publicly release it due to the possible misuses, including criminal misuses. This is a great way for the layman to get a feel for just how advanced Deep Learning has actually become.
    Last edited by ClaytonB; 05-09-2020 at 03:02 PM.

  2. Remove this section of ads by registering.

Similar Threads

  1. 'Deep state' needs to be deep-sixed
    By timosman in forum Books & Literature
    Replies: 5
    Last Post: 04-15-2018, 10:07 PM
  2. Replies: 0
    Last Post: 08-13-2017, 01:13 PM
  3. O The Deep, Deep Love of Jesus....
    By jmdrake in forum Peace Through Religion
    Replies: 2
    Last Post: 02-28-2017, 08:20 AM
  4. McCain: "Hurr, leading airplanes is like leading an economy."
    By mokkan88 in forum Reagan Library Debate
    Replies: 0
    Last Post: 01-30-2008, 08:20 PM
  5. Botnet Researcher's bias
    By rjl in forum Grassroots Central
    Replies: 18
    Last Post: 11-01-2007, 07:30 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts