Notices by Patrys (patrys@mastodon.online)

@aral @dk

But the stocks go *brrrrrr*

@dk

Surely, a formally-defined strictly-typed prompting language should suffice?

Terry Pratchett's wonderful Thief of Time contains a scene where the Auditors, god-like bureaucratic beings, try to understand art. But instead of looking for emotions and beauty, which they are unable to comprehend, they analyze the distribution of pigment on the underlying canvas, which is easy to measure and quantify. They see art as a collection of lines and dots.

It's exactly how LLMs operate. This is a post about AI.

@feld @EnfysBook I hope not. They let you control what the AI can see, and being hostile towards your customers is a poor business model (I'm not holding corporations to any moral standards here).

I think people are afraid it will accidentally learn sensitive information and possibly leak it in the future (though it’s still contained to your local device). (1/2)

@feld @EnfysBook If you don’t trust Google or Apple at that level, why use an OS that can literally do what it wants: not just through the keyboard APIs, but also through the a11y APIs, through screen reading, unrestricted memory access, etc.?

It’s one thing to limit potential information leaks. It’s another to assume malicious intent.