OpenAI engineers have been studying the causes of rare infrastructure crashes. They analyzed thousands of 'core dumps', which are detailed error reports that can help identify problems. This led to the discovery of two issues: a hardware fault and an 18-year-old software bug. The team used large-scale analysis to pinpoint these problems, demonstrating the value of this approach in debugging complex systems. --- Why it matters: This research matters because it shows how large-scale core dump analysis can be used to identify and fix rare but critical infrastructure crashes, which is essential for reliable AI system operation. Source: https://openai.com/index/core-dump-epidemiology-data-infrastructure-bug