Urgent.News

What's breaking now, across thousands of outlets.

AI

Chat-based Large Language Models replicate the mechanisms of a psychic's con

Article URL: https://softwarecrisis.dev/letters/llmentalist/ Comments URL: https://news.ycombinator.com/item?id=49775104 Points: 149 # Comments: 234

Chat-based large language models, often referred to as "AI," have been widely regarded as intelligent. However, upon closer examination, it is clear that these models lack any inherent mechanism to demonstrate genuine intelligence or reasoning. Large language models (LLMs) are merely mathematical models of language tokens, generating responses based on statistical probabilities rather than true understanding or cognition.

The phenomenon of chatbots exhibiting an "intelligence illusion" bears striking similarities to the tactics employed by psychics and their practitioners. Through a technique known as cold reading, psychics and chatbots alike create the illusion of possessing uncanny insights into individuals. By using validation statements, such as sentences adhering to the Forer effect, both psychics and chatbots generate the impression of making highly specific predictions, when in reality, these answers are statistically generic.

The intelligence illusion in chatbots can be broken down into six key stages, reminiscent of a psychic con. First, the audience is selected, a process that typically involves an already predisposed group of individuals more inclined to believe in psychic phenomena. Next, the scene is set, with lights dimmed, and the chatbot or psychic hyped up to create an atmosphere conducive to the performance.

The audience's demographic information is then gathered, allowing the "psychic" to narrow down the demographic and make statements that appear specific yet are statistically likely.

Once the audience is engaged, specific questions are asked, aiming to create the impression of uncanny accuracy. This subjective validation loop further reinforces the delusion, as the chatbot or psychic's answers are tailored to the audience's demographics and prior responses. Ultimately, the illusion of intelligence is shattered, leaving the audience convinced that they have encountered a genuine psychic or AI with extraordinary abilities.

However, the reality is that these chatbots and psychics are merely perpetuating the same age-old con, exploiting the audience's pre-existing beliefs and expectations.

Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at softwarecrisis.dev →

More in AI

Google TPU Rationing Is Not a Supply Story. It's an Authority Story.

TPU rationing at Google is not primarily a story about running out of chips. It's a story about what happens the moment real demand for a finite resource exceeds what that resource can supply…

  • Google's TPU rationing is about allocating limited resources, not a supply shortage.
  • Physical component constraints, not forecasting errors, drive the scarcity.
  • Alphabet prioritizes frontier AGI work over Cloud and other services.

Unknown `.apc/` Paths Should Not Become Hidden Instructions

Unknown .apc/ Paths Should Not Become Hidden Instructions A repository can contain a file under .apc/ without making that file part of the APC contract.

  • Unknown .apc/ paths should not be treated as hidden instructions.
  • The Portable Context Layer advises consumers to ignore unknown paths unless explicitly defined.

AI Models for Coding 2026: Which Model for Which Budget?

Anyone who wants to code with AI today is faced with a jungle of models – and opaque pricing models. The good news: in 2026, there is a suitable model for every budget and every requirement.

  • Premium models excel in complex tasks with high code quality (Claude Opus 4.8 and GPT-5.6 Sol)
  • Budget models cater to mass processing with lower costs (DeepSeek V4-Pro and DeepSeek V4 Chat)

More from Sunday 20 September →