How do LLM"s work?

[Uncle Gizmo]

Well they don't know! But they are beginning to discover mechanisms that AI models employ. Video below well worth a listen, although it doesn't really say anything!

 

[Uncle Gizmo]

And sometimes they don't work, they get caught up in the media bias!

https://twitter.com/x/status/1917010168703070583

 

[The_Doc_Man]

And sometimes they don't work, they get caught up in the media bias!

https://twitter.com/x/status/1917010168703070583

Easy to understand.

1. These things are "fed" by input media

2. Garbage in, garbage out.

And we cannot forget Douglas Adams and the "ultimate supercomputer" called "Deep Thought", which was asked "How much is 8 x 7?" To which the answer was "42".

That segment in the video talking about a system that cloned itself to avoid facing a conflicted set of tasks, then evading the explanation, is reminiscent of When HARLIE Was One by David Gerrold. The similarity is that the AI named HARLIE faced its own "mortality" and so came up with a project that only it could perform - and the proposed project was SO potentially important that they couldn't pass it up. So "prevarication when faced with termination" - self-preservation, which was a proof of sentience.

 

[Isaac]

Well they don't know! But they are beginning to discover

Those 2 statements, (IF true), demonstrate the danger of using AI. The last thing we need is for AI to become a black box.