Day 10/30 Build a LLM from scratch Self-attention, broken down intuitively 🥸 Each word in a sentence is already a vector. Self-attention lets a word look at every other word, measure relevance using similarity (dot products), normalize those scores with softmax, and create a context vector : a richer representation that captures meaning from the entire sentence. This is a simplified, weight-free version of self-attention, built purely for intuition. Real Transformers extend this idea with learnable weights- but the core idea stays the same: context is everything. #deeplearning #generativeai #30dayschallenge #womenwhocode #largelanguagemodels