Hacker News new | ask | show | jobs
by matchagaucho 266 days ago
As context window sizes increase and token prices go down, it makes more sense to inject dynamic memories into context (and use RAG/vector stores for knowledge retrieval).