The 100,000 whys of AI
The author describes frustrating arguments with fellow techies about whether it's possible to distinguish between human-written and AI-generated text, framing this as one of the most painful recurring debates in tech circles.
Background
- The article is by Michał Zalewski (lcamtuf), a well-known security researcher and author of "Silence on the Wire" and "Understanding ESP."
- It argues that AI-generated text lacks the "100,000 whys" — the deep chain of lived experience, reasoning, and incidental knowledge that shapes how a human writes. A human doesn't just produce words; they carry a lifetime of tangential facts, mistakes, sensory memories, and internal contradictions that seep into their prose.
- The piece pushes back against the common tech-industry claim that AI text is now indistinguishable from human writing — a claim often made in marketing for large language models like GPT-4 or Claude.
- This debate matters because if AI text truly were indistinguishable, that would justify replacing human writers, undermine attribution, and change how we evaluate originality. Zalewski argues that experts who actually work with text (not just benchmark scores) can still tell the difference, and that the superficial quality of AI output hides a fundamental lack of authentic cognition.