Make an AI sound like a YouTuber (LAB): Crash Course AI #8

Let’s try to help John Green Bot sound a bit more like the real John Green using Natural Language Processing. Today, we're going to code a program that takes a one-word prompt and then completes the sentence that sounds like something John Green would say. To do this we’re going to collect transcription files from Vlogbrothers episodes, do some preprocessing since John Green has a pretty large vocabulary, then we’ll set up a recurrent neural network (or RNN), train our model, and test it!

Crash Course AI is produced in association with PBS Digital Studios

Thanks to the following patrons for their generous monthly contributions that help keep Crash Course free for everyone forever:

Eric Prestemon, Sam Buck, Mark Brouwer, Indika Siriwardena, Avi Yashchin, Timothy J Kwist, Brian Thomas Gossett, Haixiang N/A Liu, Jonathan Zbikowski, Siobhan Sabino, Zach Van Stanley, Jennifer Killen, Nathan Catchings, Brandon Westmoreland, dorsey, Kenneth F Penttinen, Trevin Beattie, Erika & Alexa Saur, Justin Zingsheim, Jessica Wode, Tom Trval, Jason Saslow, Nathan Taylor, Khaled El Shalakany, SR Foxley, Sam Ferguson, Yasenia Cruz, Eric Koslow, Caleb Weeks, Tim Curwick, David Noe, Shawn Arnold, William McGraw, Andrei Krishkevich, Rachel Bright, Jirat, Ian Dundore
--

Want to find Crash Course elsewhere on the internet?

[IceMetalPunk]

The pure defeat in your voice as you read the words "totes amazeballs" was visceral. I completely empathize with you; when I first saw that commercial that emphasized the phrase "totes McGoats", I immediately had a migraine. Stay strong, Jabril!

[scm]

This is how you can fix the Runtime error on Step 3.3 (shape invalid):
In step 3.1 there is a line that reshapes the tensor prediction_vector. The problem is that the view function, which is meant to reshape the tensor, in step 3.1 is not being adjusted properly. If there is any situation that you don't know how many rows you want but are sure of the number of columns, then you can specify this with a -1. Only one of the axis values can be -1! This is a way of telling the library: "give me a tensor that has these many columns and you compute the appropriate number of rows that is necessary to make this happen".

Thus, the fix to this error is to go into Step 3.1, find line #11, which looks like:

prediction_vector = * batch_size, -1)

and change this to be:

prediction_vector = prediction_vector.view(-1, len(vocabulary))

Letting the first dimension auto-size, and the second dimension be equal to the size of our vocabulary.

[Mussul]

Wow, this video was so detailed, full of information, that it is on par with some computer science classes back in my days of university!

[janyjj]

I WOULD LOVEEEE IT IF CRASH COURSE HAD AN ACCOUNTING COURSE!!❤️️.

[hopemoore]

I haven't seen 1 through 7, but I was excited to see Jabrils as the host! I tell people to watch his machine learning videos! You're awesome, dude.

[PalaceOfWisdom]

I forced a bot to watch 1000 hours of Vlogbrothers...

[drs]

This course was well ahead of its time, even when it was made a few years from now. I love it

[Danilego]

What an amazing video! The program was great to follow along and produced some really funny results. Jabril was a fantastic host as always, I laughed so hard at "totes amaze...balls..." 5:50 hahaha so good!

[IceMetalPunk]

Thank you for introducing me to Hugging Face... now I'm just playing with it all day and alternately being very impressed and laughing my butt off.

[princessmia]

Came here from the aforementioned Wiki to add a transcript of this video :P Thanks for the shoutout!

[marksusskind]

Totes-Amazeballs? Scriptwriter_AI must've not ruled out rare words.

[TheJaredtheJaredlong]

I hope someone compiles a super data set of every word John Green has ever written and spoken online: his books, vlog brothers, brotherhood 2.0, crash course, twitter, commencement speeches, tour speeches, hacked personal emails, etc.

[Shnrue]

wait, jabrils can talk directly ???! where's the dubbing ??

[andymorrall]

If you want the AI to produce "interesting" sentences, it not only needs a model of what is possible to say, it needs a model of the listener, so it knows how acceptable they might find "totes amazeballs" and "Lilliputian".

[Majin]

AI needs courage.
Times one billion.

[Herr_Vorragender]

What the ...! Zombicorns?!
Now I really want John Greenbot to write a poem about Zombicorns :D

[Danilego]

5:10 Do I see... ZIPF'S LAW! Micheal liked that!

[PenguinsKill_l]

Now we need to see all the words that were said once in the actual videos

[actionkey]

CrashCourse
- AI examples are a good idea

[alexixeno]

Oh gawd... I feel sorry for any Ai based on V-sauce. That man is chaotic enough, let's (not) create a AI trying to imitate that orderly randomness