@yash_mittal_dev Iβve also launched my own agent on meAsAgent, and itβs honestly amazing!
Really loved seeing how it captures my personality, interests, and the way I work. Such a fun experience!
Check it out: https://t.co/eYexv9JYuV
So next time something like LLm replies, remember what actually just happened It sliced sentence into pieces. Turned each piece into a number. Looked that number up to get real coordinates. Tagged each word with its position. Then guessed one word at a time, out of 1000 of option
Ever wondered how ChatGPT actually "reads" what you type?
Here's the wild part: it can't read at all.
It doesn't understand a single word. Not one.
Here's what's actually happening behind the scenes
Step 8β Remember the order - A plain meaning-vector alone doesn't know word order.
"Dog bites man" and "man bites dog" would look exactly the same without this step. Big difference in real life though. So each word also gets tagged with its position like a seat number at a table
LLm was never taught to chat.Its entire training was one game: predict the next word- across 300B tokens Meaning came for free. Self-attention lets every word look at every other word that's how it knows bank of the river isn't about money. No dictionary. No rules. Just attention
Day 1 β Learning How AI Retrieval Works
Started from the basics and built a keyword search engine from scratch.
Tokenization β Matching β Scoring β Ranking
Next, we'll solve these limitations one by one.
GitHub: https://t.co/ZWpj7pJSOk
#AI#RAG#BuildInPublic