Post #4426035

As others have noted, and in my opinion, the recently advertised “tests” of the capabilities of several LLMs to crack computer security defenses are to a large extent publicity stunts. Simplistically: I. XYZ-1, Inc.: “Our AI is dangerous.” II. XYZ-2, Inc.: “Our AI is more dangerous than yours.” However, if those models can indeed be “tested” only by letting them out into the Internet, they begin to resemble the weapons of the so-called Star Wars project of the early 1980s, which could only be “tested” in a real war. [P.S. The official name was SDI, Strategic Defense Initiative, if I recall correctly.]