Qapla: Someone Trained a Small Language Model on an $8 ESP32-S3
GitHub / Hacker News·low signal
A GitHub project trending on HN (41 points, 15 comments) demonstrates training — not just inference — of a small language model on an ESP32-S3 microcontroller costing about $8. The interest on HN is less about the model's usefulness than about establishing the actual floor of what training hardware costs, at a moment when the dominant narrative is 474 GW interconnection queues and trillion-dollar capex. Worth watching as a reference point for edge-training claims rather than as a production tool.