AnalysisAI ModelsNovember 19, 2026

LLMs read and write executables byte by byte from a hex dump

Read original source →sgiannopoulos.github.io

A blog post tests gpt-6.1-sol (effort high) via Codex CLI 0.166.0 on raw hex dumps at ~2.4 tokens per byte, with no assembler, disassembler or execution. It identified all four test binaries, including a stripped FreeBSD ministat, and found a real bug: -c 99.5 uses critical values for ~99.8%.

1 source

More stories today

Open the live feed