Met a guy making $1.6 million/year as an LLM engineer. I asked him how he learned LLMs from scratch. He sent me the exact video that got him in. A 1 hour course on how LLMs actually work. He shows how transformers inside LLMs like ChatGPT & Claude are... actually built. I watched it last night. Halfway through, I realized LLM architecture is way simpler than they make it look. Bookmark this and read the article below. • 00:00 - LLM foundations • 04:21 - LLM tokenization • 05:43 - LLMs vector embeddings • 22:16 - attention mechanism of LLM • 43:42 - LLM multi head attentionshow more

starting with tokenization makes the later attention math click

absolutely gipp, LLM tokenization to multi head attendance everything is covered in this

This is the real cheat code to win using ai

absolutely understanding how LLMs work is 101 to win in ai

if he has sent you this course he must be good roan!

he is literally the LLM master

transformers visually explained wow

yeah LLMs visually explained

almost done watching this, i'm amazed saving it to rewatch later

I love the lectures; you always publish truly useful content

always trying to add value by sharing such highly valued videos like LLMs

Thanks for sharing this habibi

glad to add value on LLM

Youtube link of this lecture🚀

The people who understand how transformers actually work will always have an advantage over the people who only know how to prompt them.

Pure alpha, thanks for sharing

understanding how LLMs are built is truly alpha

Hey Super, build an interactive LLM Tokenization and Embeddings Explorer using D3.js. The user starts by typing a short sentence into a prominent input field. As they type, the text is immediately split into visual 'token' pills in a horizontal tray below the input. Below the tray, render a 2D vector space (a grid) using D3.js. The tokens from the tray animate into this space, taking up pseudo-random but clustered positions (e.g., nouns closer together, verbs elsewhere). The core interaction: the user can drag these token nodes around the vector space. As nodes are moved closer to one another, dynamic SVG lines connecting them increase in thickness and opacity, simulating 'attention' weights. Moving them apart weakens the connection. Include a 'Reset' button to return nodes to their original clustered positions and an 'Export' button to download a PNG of the current vector space state. The design should be a sleek dark mode with neon accents for the tokens and attention lines, ensuring high contrast and readability.

Understanding the fundamentals makes everything else easier to learn

This is the exact type of resource that actually moves the needle. The progression from embeddings → attention → multi-head is perfect. Any other underrated videos/courses this $1.6M engineer recommended?

understanding tokenization and attention changes everything so much

Once you see the actual architecture, the mystique disappears fast.

he told you something really smart, i saved this video where do you even find friends like that? haha

It's just doesn't stop there

1 hour well spent beats months of cope

the mechanics are simpler than the mystique

That attention mechanism part is where it clicks for most people honestly.

Just a one hour course? Looks like he just got luck not that he worked hard

Learning LLM architecture is not that add. The discovery of it might have been. But once the cats out of the bag it’s just a bunch of expensive matrix multiplications. What’s hard is showing up like this guy and speaking in public with the confidence he’s exuding. Thats what earns the big bucks. Knowing your scriptures and being able to evangelize others.

Nice

The best way to learn AI is still by building a strong mental model of the basics.