"We can't ask AI, it lies" vs. "Here is my superpower prompt"
The article contrasts distrust of AI due to its tendency to lie with an overly optimistic view that good prompts solve everything. It argues for a balanced perspective that acknowledges both AI's flaws and its practical value.
Background
- The article contrasts two common attitudes toward large language models (LLMs like ChatGPT, Claude, Gemini): the "skeptics" who say AI can't be trusted because it hallucinates, and the "enthusiasts" who claim they've found a magic prompt that makes AI flawless.
- "Prompt engineering" refers to the practice of carefully crafting the text you feed an AI to get better, more reliable outputs. Some prompts — like "chain-of-thought" (asking the AI to reason step-by-step) or "persona prompts" (telling it to act as a domain expert) — are known to reduce errors.
- The deeper issue is that LLMs are next-word predictors, not truth-seeking databases; they do not "know" anything and will confidently generate false information (hallucinations) when they lack the correct answer. No single prompt can eliminate this structural limitation.
- The post argues that both extreme positions are naive: prompt tricks improve reliability but don't fix the underlying problem, and dismissing all AI as useless ignores real, practical gains from careful prompting.