Видео

Can you come teach at my school?

this deserves way more views! Thank you again for the awesome videos ❤️ You are so helpful

I have been reviewing all your videos for my NLP masters program. I love how well articulated they are. Very great overview! Thank you so much for sharing this on youtube

way better explained than my professor at ucsd

Hi, that was great!

🤣🤣🤣

very clear, thanks!

awesome teacher

Professor, I just wants to say "THANKS FOR ALL YOUR EFFORT". God Bless you!

Thanks a lot for sharing your great content.

Interesting, thanks!

i dont understand anything :(

Good explanation. Thanks for posting!

Thank you, but you started explainning GD and suddenly jumped to SGD

Thanks Professor Durrett! I'm currently a undergrad student from Bangladesh, doing this course to get my prior knowledge for my upcoming thesis in NLP. Hope your course will make me grow more and more interest in NLP which seems super fun to me already. Thank you again! <3

I like this ... Thank You.

Thank you so much for such content.

In the paper by Vaswani et. al., they describe their sinusoidal positional encoding as "a geometric progression from 2*pi to 10000 * 2*pi". Any help decoding this sentence? Thanks

I watched the whole series videos. Thank you for your hard work!

👍👍👍😉

thank you...that was a very clear explanation to a complicated topic...

Best class intro ever haha

sad lols haha

That's the awesome trailer!!!!

very clear explanation! Thank you❤❤❤

In time stamp 2:05 it is implied that chatGTP is an encoder-decoder? Isn't it a decoder-only architecture?

*PromoSM* 💃

Can you please explain how adding an extra feature to the feature vector the equivalent of having a entire bias vector? Correct me if I’m wrong, but a bias vector would contain biases for each feature (or node in a NN layer), whereas adding a bias as a single new feature would be a single bias alongside all the features so could only have an aggregate effect. It seems like these are not equivalent and would have dimensionally different impact/results.

How do you generally handle misspelled words in pre-processing? Do you rewrite them as corrected words or treat them as UNK?

ahaha this is so cool. I'm excited!!

super nice :D

Dang it. Now the updates come through. I took the course a year too early 😢. I'm hyped for my classmates at least.

Great trailer! I'm ready 💪

I'm ready..

Nice trailer! I took the course 2 years ago before all this NLP breakthrough and I really enjoyed it

Hi Prof Durrett, thank you so much for releasing your course material online! I took the past version of this course last year through the MSDS program, and I'm super excited to go through the new material!

[Mask] = [your mom]

This lecture is highly intuitive . Thank You !

Thank you so much for helping me learn this concept.

I tried using a sabre fencing athlete and it thought it was a baseball player or something, probably because I used little context words.

I think the logistic function is: 1/1+e^{-f(x)} instead of e^{+f(x)}

Lol chaos

very nice :)

Hello

Hi there I am joining in Fall 2022

Hey is anyone else planning to do the entire online CS388? Looking for a study partner to improve our chances of completing the entire course and all 6 assignments.

Great lecture

You sir, are the best part of of my morning so far. The day can only get better after seeing this.

The is the blockbuster of 2020

interesting idea :)