LLMs can read and write executables byte by byte from a hex dump
Summary
The article presents a technology paper demonstrating that large language models can read and manipulate executables byte-by-byte from hex dumps, including reading, extending, and writing machine code. It details several experiments where LLMs analyze real binaries, patch code, and even analyze malware, illustrating potential capabilities and implications for reverse engineering and programming with AI, as well as security considerations.