Raw LLM Responses

4:20 This is a child's understanding of current AI technology. Associative Procedural Engines (LLMs/MMMs) are not PREDICTIONS, they are probabilistic associations. It is not predicting humans, particularly not individual ones, but finding the most likely next token, which somewhat changes depending on the input and output format. This is a pure I/O machine that people give mystical powers of thought and foresight. It cannot think, it cannot understand, it cannot even conceptualize. It can provide most likely categorization, which, provided a recursive mechanism, can imitate the outcome of a pattern, in linguistic, auditory, graphical, or other implemented formats. This it's why system prompts, rules, context alignment, and every other change to weights and biases doesn't achieve expected goals; the reason we use these systems is that humans are really, really bad at understanding even small systems and our stupid human brains have spent all of human existence understanding that linguistic patterns are indicative of internal processes, so we transitive property ourselves into believing a vast Plinko machine is basically just Star Trek's Data. There is SO much damage these applications are doing exactly because people are pretending it's going to kill us all because it "wants" to, and dancing into us killing ourselves by giving extraordinary power to a random number generator and assuming it'll turn out fine.