Training a 4B model to produce 81% faster query plans than Postgres

Rohan Bansal, a Y Combinator-backed startup, has been working on a new open-source project called QORL, which aims to train a 4 billion parameter model to generate more efficient query plans for PostgreSQL. The model has been trained on a dataset of 100,000 queries and has achieved an 81% improvement in query plan generation speed compared to Postgres. QORL uses a combination of techniques from reinforcement learning and self-modifying code to adapt to different query patterns and optimize performance. The model can be trained on a dataset of queries and can be fine-tuned for specific use cases. The project is still in its early stages, but the results show promising potential for improving database performance.