Hacker News Reader

That's the difference - they have well defined and bounded errors. LLMs do not.

Which is a notably different argument than whether they can lie to you.

Why are we skipping over the miscalibrated, defective, or ill-documented devices bit though? Those also all have arbitrary error.