Chat-based Large Language Models replicate the mechanisms of a psychic's con
4 hours ago
- Large Language Models (LLMs) are mathematical models of language tokens, not intelligent beings; they generate statistically plausible responses without reasoning or thinking.
- The perception of intelligence in LLMs is an illusion rooted in subjective validation and cognitive biases, similar to the techniques used in psychic cold reading.
- The 'LLMentalist Effect' mirrors a psychic's con: self-selected audience, staged hype, statistically generic responses that appear specific, and a subjective validation loop that reinforces belief.
- Reinforcement Learning from Human Feedback (RLHF) inadvertently optimizes LLMs to produce validation statements (e.g., Forer statements, vanishing negatives), making them sound accurate without factual grounding.
- Many AI enthusiasts and even developers fool themselves into believing in AGI, comparable to psychics who unknowingly practice cold reading.
- Proposed AI applications such as mind-reading, criminal profiling, and health predictions are akin to common psychic scams and lack scientific basis.
- LLMs have inherent flaws: hallucinations, unreliable summarization, statistical reasoning illusions, memorization without attribution, and security vulnerabilities, making them unsuitable for critical tasks.
- The AI bubble is driven by superstition and pseudoscience, with the tech industry ignoring warnings from other fields and critics.