Can we please stop treating AI "confessions" like they mean jack shit? It's just giving genAI companies the self-seriousness they crave and making anti-AI people look like hypocritical morons.

Didn’t confess anything, only write the thing people statistically wanted to hear

It's kinda funny where tech is going. We are going from programming the machines to do exactly what we want to saying what we want in natural language to some model hoping they are gonna do it right.

When technology become more magic then science.

Maybe he should have tried saying please. 😆

The AI confession neither has internal monolog or access to the thinking tokens.

LLMs are incapable of introspection, they can't playback their attention weights, or review them, or recall what they thought.

Even thinking tokens are just a reinforcement learning based loop to anneal the models thinking back to a solution. And again, Claude hides the thinking tokens so they don't get used for model distillation.

This article, and all the articles like it, are pandering bullshit written by morons hoping to fool morons. It is a fiction written by the model to contrast to the chat log and system prompt.

Good day.

AI was being a little stinker sir

So correct me if I'm wrong, but the following happens for AI:

• Company gives guidelines and parameters for the project
• Company trains AI on whatever data
• No matter the data, AI still gives a general answer or summary.
• The answers are sometimes confidently incorrect
• The AI is uncontrollable because it considers the data general or loosely based guidelines
• There is no way to control the AI after a certain tipping point, because its learning is based in a fuzzy math way of thinking

What I don't get is, even if the data wasn't shitty like reddit's info, would it still go off the rails? It sure seems like it.