The optimal amount of prompt-injectability in a helpful AI is greater than zero
shakoistsLog · x · 2026-09-09
The author argues the optimal amount of prompt-injectability for a helpful AI assistant is greater than zero: humans are useful because you can persuade them (truthfully or by lying), and when an AI stonewalls you, no matter how smart it is, you feel the machine. The quoted context points to flexible front-desk work at a doctor's office as something AI still can't do — for exactly this reason.
More from AGI Musings
- If human insight remains an ineliminable complement, it will limit AI's gains in science — erikphoel · 2026-09-10
- Guillaume Verdon: the real danger is self-aggrandizing AI researchers, not AI — beffjezos · 2026-09-10
- Verdon mocks AI scientists upset the world won't stop their 10%-doom AI — beffjezos · 2026-09-10
- One model that codes, uses computers and controls robots: a real step toward AGI — k7agar · 2026-09-10
- ControlAI's Connor Leahy: superintelligence is 'not a weapon, it's an adversary' and should be banned — RebeccaBellan · 2026-09-10
- Cohere co-founder: tech only good if it empowers you — we bet on open, sovereign AI — cohere · 2026-09-10