Language models · 10 min read
What Are Word Embeddings, Anyway?
Meaning is not typed in by anyone. It falls out of a prediction game played over billions of sentences, and the leftover geometry is what we call an embedding.
Language models · 10 min read
Meaning is not typed in by anyone. It falls out of a prediction game played over billions of sentences, and the leftover geometry is what we call an embedding.
The algorithm behind GPT's tokenizer is called BPE, byte pair encoding.
We use LLMs in almost every part of our life.
A webhook is a lightweight, event-driven mechanism that allows one application to automatically send real-time data to another via HTTP (typically a POST request) immediately when a specific event occurs.
A news aggregator — like Google News, Apple News, or Flipboard — is a digital service that continuously collects news articles from thousands of publishers around the world, processes and organizes them, and presents them to users in a unified, scrollable feed.

Kishan Kumar
Senior Software Engineer · PhonePe
Senior software engineer at PhonePe. I write long explainers on distributed systems, language models, Java and blockchains.
I've spent over six years building systems at scale. At PhonePe I lead the logging charter, which handles around 65 TB of logs a day.
Read more →