APNIC researchers release PCAP-LM for AI traffic reading

Researchers have developed PCAP-LM, a novel packet-capture representation method designed to help large language models read and analyze entire network traffic files in a single prompt. By rewriting conversations into compact lines of symbols called PacketGlyphs and applying run-length and byte-pair encoding, the format reduces standard text outputs by about 800 times while retaining 99.3% accuracy on basic forensic questions. Open article →