The LLMentalist Effect (2023)
- ID
- 26787
- Status
- summarized
- Published
- 20 Sep 2026, 8:20 PM
- Fetched
- 22 Sep 2026, 3:24 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://softwarecrisis.dev/letters/llmentalist/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 6.0
- Created
- 22 Sep 2026, 3:27 AM
- Tags
- Audience
- developersai_ml_learnersai_agent_userssaas_founders
What happened
Baldur Bjarnason argues that the perceived intelligence of chat-based LLMs is an illusion driven by the same mechanism as a psychic's cold reading—statistically generic statements (including Forer-effect validation statements) that feel personally specific to the user. He claims there is no reasoning mechanism in LLMs, only mathematical token prediction, and that many proposed use cases now look like 'borderline fraudulent pseudoscience' to him.
Why it matters
If you're building products on chat-based LLM interfaces, this frames a concrete risk: users may attribute reasoning that isn't there, which means UX and product claims need to be designed so users don't over-trust outputs. Evaluate whether your agent or chatbot is relying on generically validating phrasing rather than genuinely task-specific accuracy.
Discussion angle
Where in your own AI product UX are you accidentally amplifying the 'cold reading' effect, and what would change if you stripped out generically validating language from model outputs?