News item: Amazon warns its developers not to use ChatGPT as a search engine or problem-solver. AI learns from every conversation, which means that a conversation including trade secrets will automatically become part of what AI knows. Amazon’s counterspies have noticed some trade secrets showing up in their tests of ChatGPT.

This is the inverse of the standard ‘marked bill’ method. Publishers sometimes include peculiar names on maps to spot copiers, and spies sometimes include recognizable info in their radio or web communications to spot interceptors. In this case the developers are accidentally including marked info, and the company’s own spies are spotting the marks.

But below all the spy stuff, developers who use AI as a reference are stupid. Experenced programmers can handle typical situations without help, but we sometimes require peculiar answers to peculiar problems. Services like StackOverflow bring programmers together to answer peculiar problems. (Most of the answers are snarky and useless, but occasionally you strike paydirt.) AI jumbles up relevant info to create an average or plausible answer. The result is sometimes just plain wrong at the level of bad arithmetic or nonexistent keywords. Even if you can edit out the simple wrongness, you shouldn’t have to search for an average answer.

Also below all the spy stuff, people who work with copyrighted or NDA’d info should know what’s protected and what’s not. For instance, I occasionally use some of my courseware animations to discuss a neurological point here on the blog. I’m careful to avoid the copyrighted form of the animation. I look at it from a different angle or omit some pieces. Amazon programmers shouldn’t be outputting trade secrets in ANY non-private web situation. If they really think AI is private, they’re unbelievably stupid.

= = = = =

Later: This Reddit conversation helps to explain the situation. There’s a generational pattern that I hadn’t noticed. In general, people under 25 have ONLY used iPhones, not real computers. Everything on an iPhone either works totally and easily or doesn’t work at all. There’s no need to understand file folders or memory size or ANY of the underlying details, and there’s no way to troubleshoot. Oldies like me, ranging down to about 40, started with DOS or early Windows on desktops, and never gave up the desktop. Even if we weren’t serious programmers, we had to know what and where and how big a file is.

Another prominent theme in the Reddit discussion is cybersecurity. Schools can’t teach the inner details even if they wanted to, because TRUMP_PUTIN_HACKERS might learn something. This ties closely with the trade secret stupidity. If you can’t even learn where and why REAL security is needed, you’re lost.