How LLMs actually work:

Video credits: @3blue1brown Full details:

Maybe credit the original creator here?

This brought back memories of the Discrete Mathematics and AI course 😅

Wow what a nostalgia

How about we create a culture on X where we credit our hardworking creators video is from : of @3blue1brown ,if you find his work valuable please support him : @grok expand on Grant Sanderson and his work

Did @3blue1brown make this?

Great video

Do you think people who doesn’t know how it works before seeing this videos , understand something ?

And it's crazy to think all that is driven by electrical current. We brought electrons to perceptrons. What a legendary civilization humans are!

its insane how the human mind does something more similar and powerful to this

3B1B?

Thank you, hugging models 😛

🤗

Love how the replies called out the 3b1b credits - his visuals really make complex topics click for devs

Great visualization 💯

Who wrote this

Really neat explanation.

This is not an explanation this just a remix of the great 3b1b video about transformer architecture.

They make the video fast so people can keep not understanding anything of that. Someone must explain how LLMs work in a way a child can understand.

Amazing

This is a better way to learn about LLM.

Good to know

@jingolin2

Great one

That an amazing illustration of an llm working

Power of Linear algebra !! 🤩

He usually does great work but this one was too ambitious. The energetic music captures the overwhelming confusion.

you mean how they actually DONT work 🤣😱👀

@HuggingModels, understanding transformers is key to grasping how LLMs perform their tasks effectively.

Made for people who already know how LLMs work.

Please credit @3blue1brown for his work as opposed to ripping it off.

🐮b

@xviddownload

And all of that you see, is the transformation of current token embedding into the next one, it's just linear algebra with non-linear functions.

Just reminds me of a giant inefficient brain

@grok what is LLMs

And that iteration is repeated for EVERY WORD (token) until the full response is complete. That's why processing so many prompts each second can be achieved only using graphic card which are designed to process insane amount of computations at blazing speed.