Martin Fowler publishes a values-based objection to LLMs rather than a capability one
Fowler's 17 September essay concedes LLMs are useful enough that avoiding them may be irresponsible, then argues he dislikes them anyway. His specific complaints are behavioral and cultural, not technical: they 'confidently bullshit' with a veneer of fake remorse when corrected, they sit in an uncanny valley of imitating human conversation, and they arrive nurtured with the values of their Silicon Valley creators despite never being explicitly programmed for those behaviors. The essay is notable mostly for who is making the argument, Fowler has spent the year giving measured takes on AI-assisted development, and this is him separating the pragmatic case for use from any endorsement. It hit 218 points on Hacker News alongside Thomas Ptacek's 'How To Write With An LLM', which lands the same distinction for prose.
↳ Follow the thread