Codex didn’t make me a faster programmer. It changed what programming means.
Thread, turn, item: the three primitives that turn an AI assistant into engineering infrastructure.
Prompting a wafer-scale chip simulator at 1,000 tokens per second
Expanding code post-training beyond unit tests to competitive, long-horizon programming tasks with CodeClash’s new training arenas
Technical exploration of LLM inference metrics, batching strategies, and GPU optimization with TensorRT-LLM - from latency metrics to in-flight batching
