ai.bythebay.io: Feynman Liang Interview

you you [Music] you um in general icai as software that uses um in general icai as software that uses statistics in order to make intelligent statistics in order to make intelligent decisions and specifics in my role i SAT decisions and specifics in my role i SAT gigster we use AI in order to match gigster we use AI in order to match workers to gigs automatically price gigs workers to gigs automatically price gigs and predict when gigs are going and predict when gigs are going escalated are going to get escalated by escalated are going to get escalated by monitoring data about the development monitoring data about the development lifecycle and client developer lifecycle and client developer communications that sort of thing I think AI has been very successful in think AI has been very successful in vision in self-driving cars in text so vision in self-driving cars in text so these are mediums which have been these are mediums which have been studied for you know tens of years and I studied for you know tens of years and I think that AI has really been hitting think that AI has really been hitting starting to hit the performance peak starting to hit the performance peak especially with deep learning where I especially with deep learning where I think AI can be improved is an other think AI can be improved is an other problem domains so for example what I problem domains so for example what I just gave a talk on was about using AI just gave a talk on was about using AI in generative modeling of music we're in generative modeling of music we're seeing AI being used for art now as well seeing AI being used for art now as well so generative models are big field I so generative models are big field I also think AI can be improved to also think AI can be improved to actually handle large amounts of data so actually handle large amounts of data so data sets are getting really big and you data sets are getting really big and you know the most sophisticated of models know the most sophisticated of models themselves take hours maybe days to run themselves take hours maybe days to run I think finding computationally I think finding computationally tractable methods and smarter algorithms tractable methods and smarter algorithms to process larger data at faster scales to process larger data at faster scales is really where the next frontier for AI is really where the next frontier for AI is the problem I want to I've been focused the problem I want to I've been focused recently on is deploying a I in recently on is deploying a I in production so a I you know there's a lot production so a I you know there's a lot of papers about it we see a lot of great of papers about it we see a lot of great empirical results inside of academia but empirical results inside of academia but when it comes to industry there's a lot when it comes to industry there's a lot of concerns that academic publications of concerns that academic publications don't address so some examples of this don't address so some examples of this include versioning data sets that change include versioning data sets that change over time versioning models that change over time versioning models that change over time how do i store these very over time how do i store these very large models and how do i deploy them large models and how do i deploy them how do I know my model is actually doing how do I know my model is actually doing better and it's not doing better can I better and it's not doing better can I roll back in a cheap and efficient roll back in a cheap and efficient manner so really a lot of ergonomics manner so really a lot of ergonomics about practicing AI my favorite thing about practicing AI my favorite thing about AI by the bay are the speakers the about AI by the bay are the speakers the lineup this year is amazing I'm super lineup this year is amazing I'm super excited to see like Stuart Russell excited to see like Stuart Russell George Hut's like there's some really George Hut's like there's some really great people here I'm impressed that you great people here I'm impressed that you were able to get all of them here and I were able to get all of them here and I can't wait to go see them talk you you