Weeks after OpenAI released GPT-2 in 2019, a lone French engineer published a paper using the same Transformer architecture for something nobody was hyping. Compressing files. That engineer was Fabrice Bellard, who had already written FFmpeg and QEMU. Today his compressor sits at the top of the Large Text Compression Benchmark, a contest AI researchers have run for 20 years as a measuring stick for machine intelligence. The connection is one almost nobody traces. To compress text, you predict…