Study argues LLMs use cold-reading tricks similar to psychic cons

A new analysis published on softwarecrisis.dev draws parallels between how chat-based large language models respond and the deceptive techniques used by psychics and mentalists. The piece argues that LLMs, like cold readers, generate responses that feel personally accurate by using vague, statistically likely statements. This creates an illusion of deep understanding or knowledge when the system is largely pattern-matching and inferring from context. The author suggests users may be misled into overestimating AI capabilities due to these psychic-style conversational dynamics. The article sparked discussion on Hacker News, signaling growing interest in critically examining how LLM outputs are perceived and trusted.