thank you , it's an amazing explanation

didnt explain why clipping, min or anything for the loss

asians: tell me something which chatgpt doesn't explained you. NR Asian: lets make a video explaining dumbness.

excellent explaination

I think there is a miscalculcation at 8:26

You deserve a tons of like!!!

Love the quizes👍 and this teaching format🔥

مرسی.

Ive been a sub since I was in undergrad several years back, you popped into my feed for whatever reason again after years. I'm glad to see you're still putting out gold, good for you man. Edit: I was halfway through the video when I commented, and now I finished the video. Those clarifications that pop up from under the screen are the best thing since sliced bread, wow

God Level Explanation. Thanks Man !

This is such a great video, thank you!

I don't understand why you don't have much more views and engagement. Your videos are some of the best explanations out there. I've sent my students to your channel multiple times. Not a great timeline where virality reins over veracity. Amazing work.

frustrated. i'm tired of having to talk to an AI that has a short context and short token window, can't remember what we were talking about 2 minutes ago, and isn't aware of what it's saying.

I want to be Optimist but now I'm Terrified. Here is my take. Before the gpt era at least in the domain of NLP, for every downstream tasks there are a lot of model training happening like NER, sentiment detection, grammar correction etc etc. Any product which needs these task to be done they need a team of NLP engineers. But post GPT era everything is just an API call away. You don't even need to know the ABC of NLP still you can perform any of these task. Now some might say that we have jobs in LLM Engineer or prompt engineer well my answer to them is Please grow up(for me prompt engineer is just like content writer job). Thankfully till date there is no single image model available which can do everything. People say Ai will effect jobs like customer care, data entry etc but I believe the first job it will effect is the job of a AI research engineer.

Optimistic as an AI Engineer because I am always hooked to latest innovations. Terrified because human intervention in AI applications is decreasing, and I fear losing my job

Optimist ofc . Next , with code pls

Ahhh, the assholes responsible for the hard college calculus life of no real life everyday application.

I love the maths and science behind LLMs or ML/DL in general. BUT, how can we supposed to be optimistic about it when we know that aim is to reduce labour and cost so that the ultra rich get richer. Look at the new self driving taxis in LA, if it keeps like that do you know many families and children will pay the price. And don't give me the crap that AI is to reduce boilerplate code for developers or increase productivity to focus on creativity, that isn't the goal, investors care about cutting costs. Look at the dock workers who got replaced by machines. Not only they weren't paid enough, now they're paid nothing. I'm not against technology, but I'm against greed. Look at the numbers between the fifties and now, and AI will be the final blow.

terrified...

Really enjoyed it! Informative and funny😊👍🏻

Man, your content on transformers really helped a lot. Keep making content like these <3

Explained well sir!!

it was a good video - until you got to quiz time and decided to try to click your tongue and make incredibly annoying jeopardy sounds. that cost you a like and a subscribe

1. A. As, Frank uses the Q table to decide the best action in each state by selecting the action with the highest Q value. 2. C. Past experiences are being stored in a replay buffer, while training we pick random samples in order to break the temporal correlation inherent in sequential data so as to allow model to learn from a diverse experience. 3.C. Target network helps to ensure that the targets do not move dynamically during the learning updates, providing a fixed point for the optimization process.But yes after some episodes ,you need to make sure that target network is synchronized with the main Q-network.

AI is just born 🤣🤣

great work! 很好的视频！

Try to add pre requisites. Begineers might watch this and get nothing. Might be good for reach, but seems like clickbait.

Best explanation for beginners like me in time series forecasting ❤

i have a question so the vanilla autoencoder is completely useless we should stay away from it and instead we use regularized autoencoder as the default autoencoder?

Hey Ajay, your explanations are nice, like always, though I didn't get how the target network can represent the ideal dostribution if we have no direct access to the true underlying process. Also I'd avoid the quizzes, they break the flow

Thank you Ajay! Lovely video. Very informative and helpful. What I truly like is the fact that you introduce relevant research papers in the videos and explain the concept thoroughly! The quizzes are also great! Keep up the good work and thank you once again!

is he Indian?

Finally, a great video that explains the entire training process so clearly and effectively! Thank you so much for this. Would be great if you can create a video on Direct Preference Optimization(DPO) as well. :)

Great Video

The diagram is great..but you should explain the code also corresponding to the diagram

2. B

Wow, you are really good at explaining things. Thank you!

Q1. B

Using BERT with an entropic recursive neural interface environment would be quite effective.

Corrections required in the video: In the qkv_layer, the bias should be set to False. The "values" variable should be permuted to (0,2,1,3) before reshaping.

Hey the flow chart diagram for Informers is very useful please share it in a pdf format or snapshot in github please, i kinda need it for my project on informers

i really appreciate you. I'm 56 autistic af and don't code. this is so beautifully structured and helpful. you are a great teacher.

amazing.

I cannot use SHAP and LDA at the same time, an index error occurs. Can someone who knows help me?

This was brilliantly explained. Thank you!

6:56 actually you would use the variance instead of the standard deviation, so in the layernorm formula it should be sigma^2 in the divisor. and the epsilon is missing to prevent zero division

How to study Probability Theory for Deep Learning ?

Please expain wotdnet also...

First Pass => D (All of the Above) Second Pass => B (Reduces computational complexity by focusing on subset of queries) Third Pas =D (10 for original transformer; 6 for informer)

Amazing video, you are a genius, please keep releasing new videos. you are the most handsome and smartest person in the machine learning field.