Andrej Karpathy on Software 2.0 at Tesla

Please support this podcast by checking out our sponsors:

GUEST BIO:
Andrej Karpathy is a legendary AI researcher, engineer, and educator. He's the former director of AI at Tesla, a founding member of OpenAI, and an educator at Stanford.

PODCAST INFO:

SOCIAL:

[LexClips]

Guest bio: Andrej Karpathy is a legendary AI researcher, engineer, and educator. He's the former director of AI at Tesla, a founding member of OpenAI, and an educator at Stanford.

[Friedrich]

Interesting way of thinking about it! Definitely see "Software 2.0" becoming a thing, but software 1.0 will stick around. Just think of all the situations where you have either a) a really simple task - here it's easier to just write the code yourself - or b) a task where it's crucial to know deterministically what's happening.

[erikpaulsen]

The model complexity is so high it does not scale with multiple programmers coming and going. So they use something like the Google Translate engine to search for patterns in massive datasets and then compress that into a logical entity that transforms live input into dynamic instructions. The problem then becomes that Tesla cars and Tesla bot non-cars behave like english and russian in Google Translate.

[rodolfoblasser]

This is mostly what edge computing is all about. What a time to be alive 😎

[zofe]

This solves the classical problem of translating UML to binary-image of a propertly working-code.
"AI" solves for it by means of translating the system-requirements and its ensueing architecture into an abstract-terrain surface, around which a Jacobian is traversing under the guise of statistical-multiplexing.

[bmxdima]

He's so smart and clear. Very good to listen to

[robertlis]

Standard ifs and whiles are much more predictable easy and cheap for a vast swaths of problems. Sure, with how cheap AI is getting more software will include components that use learning, but these is different than “nets taking over”.

[theburnetts]

But doesn’t the neural net have to be written in classic Software 1.0?

[himmura]

Definitely different from Kanye interview. 😬

[kiddd]

Is it surface pro near Lex? Which model?

[mattymattffs]

He's right and wrong at the same time. When we look at the output of much of the ML being employed, it's a lot of fucking shite for the most part. I don't think code is going away any time soon.

[luncui]

Anyone can suggest a software 2.0 course? Thanks🎉

[raz]

Software 2.0 Dev is the new real developer. Todays devs will be in 10years like a Computer User in the 90‘

[alfredoma]

So more or less how you train a human in school. Give it a lot of examples, give it a test to determine progress and then pass and deploy or feed it more examples

[ascgazz]

This guys not good at explaining things.