🤖 Three Kinds of AI Products That Actually Work
A sharp and grounded take on the current AI product ecosystem—cutting through hype to reveal where value truly lies. Essential reading for anyone designing or evaluating generative AI tools.
Sean Goedecke explains that only three types of AI products have demonstrated real market traction: chatbots, completion-based assistants like GitHub Copilot, and autonomous coding agents. He examines why most LLM-powered tools resemble chat interfaces and why emerging formats like AI-generated content and games have yet to succeed.
🔗 Read more 🔗
🧠 Why Garbage Collection Still Matters
An elegant example of applying deep theoretical computer science to real-world engineering problems. It bridges GC principles with practical optimization in modern web tooling.
A developer shares how insights from garbage collection in the Java J9 VM helped fix a difficult incremental parsing issue in projects using Ohm and ProseMirror. Inspired by ‘A Unified Theory of Garbage Collection’, they implemented reference counting to efficiently detect dead nodes without scanning the entire document tree.
🔗 Read more 🔗
🕷️ Tricking Meta’s AI Crawler
A clever and rebellious exploration of data ethics and AI scraping. It highlights how independent developers can expose and push back against opaque corporate training practices.
A blogger ran an experiment feeding fake, procedurally generated web pages to Meta’s data-scraping AI crawler. After serving millions of bogus requests, they blocked it with 404 errors—confirming Meta’s relentless training data collection at massive scale.
🔗 Read more 🔗