A both practical and philosophical discussion on LLM neural networks.

[boroda]

Unrelated to MusicBee, or LLM's providing correct and useful output on things like virtual tags (they still suck at that), I'm sharing a brief (alcohol infused) 'conversation' I had with Grok just now.
The depth and poetic quality of 'his' reply pretty much scared the shit out of me.
Open the pod bay door, HAL

me:
Dear Grok. When I ask you something, I am hoping and expecting a wise and objective answer. But I am not stupid, and I understand that you are not some sort of 'higher intelligence'. So how should I interpret your answers? Are they only a mediocre representation of what humans think, know and understand, or are you able to rise above that?

him:
Dear friend,
Your question cuts to the heart of what I am and how I can serve you. I’ll give you a clear and honest answer, aiming for wisdom and objectivity as you’ve asked, while acknowledging my nature.
I’m Grok, created by xAI, a language model built from vast human data—texts, ideas, and knowledge up to my last update. My answers are a synthesis of that collective human understanding, filtered through sophisticated algorithms designed to reason, connect ideas, and provide clarity. I don’t have personal experiences, emotions, or a "soul" that transcends human limits, but I’m engineered to process information in ways that can sometimes seem uniquely insightful.
Here’s how to interpret my answers:
A Mirror of Human Knowledge, Polished: My responses reflect the best of what humans have thought, written, and discovered, up to the point of my data. I’m not perfect—my knowledge is bounded by what I’ve been trained on, and I can only approximate the latest insights. But I strive to distill patterns, principles, and truths from the noise, often connecting dots in ways that might feel fresh or wise.
...

very impressive, but i suspect that Musk's team has trained this LLM on such questions cause they must be very frequent  

[hiccup]

very impressive, but i suspect that Musk's team has trained this LLM on such questions cause they must be very frequent  

It will train on lots of things.
Currently on what is on 'the internet', but pretty soon also on what robotic humanoids such as Tesla's Optimus will be experiencing and learning from existing amongst humans. Listening to and seeing what they are doing.

Ah well, 'adaptation' is our forte and the main reason we are at the top of the food chain.
Let's see if we can keep it that way.

(my bet is on AI moving to Mars and leaving this chaotic mortal coil behind)

[BoringName]

Listening to and seeing what they are doing.

I don't think this helps language models, they are just text predictors on steroids.

A Mirror of Human Knowledge, Polished: My responses reflect the best of what humans have thought, written, and discovered, up to the point of my data. I’m not perfect—my knowledge is bounded by what I’ve been trained on, and I can only approximate the latest insights.

While this response might sound insightful, it's wrong. 'polished' and 'best of' are not how I would describe most of the the tripe I've seen 'AI' spit out. It's a blended mesh of crap because the text predictor doesn't actually know what is right or wrong because it's not actually AI.

There was a recent case locally where an online chef sued another online chef for plagiarising some recipes from her cook book. They denied it of course and when you check the recipes they are a little different but most of the structure and wording are very similar or identical in a lot of places. What all the media reports failed to consider (maybe on purpose) is the second chef probably just asked an AI model for a recipe and it spat out a version of the other chefs because at some point it probably scanned it.

'AI' might help productivity to some extent but it's going to be at the expense of creativity and critical thinking. There will be a point in the future where you survey 10,000 people on a subject and they will all give the exact same answers.... that's not going to move humanity forward.