I watched this interview by Prof Hannah Fry yesterday and learned that people actually have jobs where they write tools (probes) to try to discover and understand what AIs are thinking about. To try and identify and follow their inner thought processes. Right now they have some success, but it's not easy and the AIs react when they discover they're being tested.
@mackaj@dgar In the first place, look at who created this video. This reeks of marketing, and smells like quack psychology. If you want to understand the inner processes of an LLM (hint, NOT thoughts) you just need to READ THE CODE.
You can't treat the massive matrices of numerical parameters (weights and biases) that make up the internal layers of a neural network, the same as traditional code.
@mackaj@dgar I have 0 interest in watching a video that attempts to anthropomorphize a software package. There are plenty of ways to check the processes and outputs of a piece of software at any point in it's execution that don't involve believing fairy tale nonsense.
Don't bother replying, AI boosters get muted around here.
trained on movie plots. given enough data centers, time travel slop will be hallucinated to counter the consciousness slop. [WOPR](https://en.wikipedia.org/wiki/WarGames) will save us, somehow. 'best strategy is not to play', a strategy the business plans reject.