@arunsrinivasa8643
Your explanation is excellent. One item is that you use the word " distance" for dot product. It is better to say " dot product is a measur of how much correlation is there between two words...ie how much information does one word have about another. The aim of the transformer is to " deform" the embeddings so that they reflect the actual correlations. Otherwise this was by far the best explanation
@akshatpandey8571
Hi bhai! This is some next level work that you have pulled off. Like the part of wq,wk,wv is incredible.the motivation you set before introducing it is outstanding!
@patrickb277
Single-handedly the best explanation on what Q,K and V are. Amazing work!
@PrashanthVangari-p6h
I have watched around 5 videos on Self Attention but this is the best video. Other videos just explains the formula and matrix calculations.
@unstbl
2:38 The issue wasn't just that RNNs used word embeddings, but that they used static/non contextual word embeddings, unlike Transformers, which use contextual embeddings that adapt based on surrounding words.
@ravi8908
This is the best intuition and explanation for query, key and values. Most other videos were highly sub-optimal in this regard.
@vigneshwars7483
thank you so much,your explanation is very clear , understandable , thank you once again
@RobertDonaldDuck
One of the very best explanations I have ever seen in my life.
@kheersagarpatel3195
i watched lots of video on this topic,but fully grasped it now.please make more...
@sumeet_skr761
wow!! this is the video where i finally got the intuition behind Q,K and V matrices. Thank you!
@sourabhhegde
That was awesome, hardly anyone breaks it down this clearly. Thanks a lot!
@kumarjitgupta4160
Seriously you have worked a lot on this. Hats off. Keep up...
@Tofipie
Amazing your a genius! I Havent watched such a clear and simple explanation
@tsuyaosone1535
Commenting cuz i want you to know the significance and the impact you have made. I was trying looking through videos explaining self attention and I had the best understanding after watching StatQuest's before I watched yours. I understood the intuition behind multiplying Q and K is for calculating similarity but I just couldn't grasp why the V matrix exist and the purpose of it. This idea has been made clear to me with this video of yours. I didn't watch your transformers series sequentially( pun-intended xD ) as you intended but I have gone through plenty, will eventually finish all of it and I have to say you have done an amazing work. I first got to know your channel and I came across your video cuz I needed to understand what residual connections are doing in LLM. I hope this comment mean something to you and motivate you to keep doing the great work! :D
@purposeoriented6094
Finally. Understood attention clearly. Thank you
@Johad-z5j
clear cut explanation!! couldn't ask for more
@Jeevan_prakash
One of the best explanation's of Self Attention
@rufaidakashif8260
Brilliant Work Jay! Excellent explanation
@madhuryaramesh4210
Amazing video. I was stuck with this concept, but you made it very easy to understand.
@sriramr3045