def attention(Q, K, V):
    scores = Q @ K.transpose(-2, -1)
    scores /= math.sqrt(d_k)
    return F.softmax(scores) @ V

What will we
learn today?

A personal AI learning library

一个个人 AI 学习知识库

Explore

Library / 知识库

Newest

How AI systems can be trained to be helpful, harmless, and honest through self-improvement guided by principles

Deep dive into the product decisions behind Cursor's meteoric rise

How Perplexity rethought search with AI-first UX patterns

Breaking down feature-based reinforcement learning with clear examples

A curated reading path from RLHF basics to constitutional AI

New research on understanding what happens inside neural networks