karan

deep neural networks

[karan @karancl]

notes from 18.06 by prof gil strang. this turned out to be the sole reason i pursued deep learning. you're still early. learn from prof strang. the lectures are pure joy.

[karan @karancl]

your epistemic originality will forever be your comparative advantage over ai.

[keshav @keshavchan]

there are going to be some moments this year, next year, the year after that when you will lose all hope. you will convince yourself that there isn't anything that you can do that AI can't. when this does happen, i want you to remember these words. stash them. you are far more versatile than you think you are. your identity is ductile. you can do new things and you can do them well. the fact that something has been maximally infused by you adds a certain ring to it that no one can. that ring is precious. that ring is yours. and that ring will always, always be invaluable. your goal moving forward, if you are looking for a north star, should be to do things only you can. triple down on your points of view. craft narratives that you like. make flavourful things. the world is kinder than you think. it will reward you generously if you give it what it truly needs. if you pay attention to its late night whispers, it is crying, yearning, longing for more of you. it needs you. desperately.

[karan @karancl]

@neuronomad_ thanks man

[karan @karancl]

career update: started as an intern at an iit, will be working on text to image models for the summer

[Jino Rohit @jino_rohit]

new in-depth blog post for "Collective Communication for Multiple GPUs".

this blog should help you understand how communication happens when you scale from a single GPU to muliple GPUs, how to reason about sizes of data you share across the different GPUs and the different collective primitives like broadcast, scatter, reduce and its variations.

id recommend this for anyone starting to learn about how to start scaling your experiments to multiple GPUs and choosing which algorithm is more appropriate and which operation is computationally more expensive to run.

blog link in comments
(bonus: lots of visuals!)

[karan @karancl]

@tm23twt thankyou!

[karan @karancl]

@indrajit_tweets thanks king

[karan @karancl]

@JasgunChandnani thankyou!

[karan @karancl]

@Jessika_almis thankyou!

[karan @karancl]

@ThedatagGuy thanks man

[karan @karancl]

@X_Ibyte thanks g

[karan @karancl]

@jino_rohit thanks king!

[karan @karancl]

@Pseudo_Sid26 thanks man!

[karan @karancl]

@jerkeyray thanks man!

[karan @karancl]

@sharpeye_wnl thanks king

[karan @karancl]

reminder that you only have to win once. it literally takes just one win to cancel all past failures.

[Dalton @Dalton254_ke]

With all these AI Agents around why isn't work reducing?

[karan @karancl]

@jino_rohit @Alibaba_Qwen big win, congratulations man!

[karan @karancl]

@C_Kumar_Kartik yes

[karan @karancl]

guess the architecture

[karan @karancl]

link https://t.co/k5KQSsa5JJ

[karan @karancl]

cool watch explaining neural networks as energy minimisers