Add 'can't write professional emails' and the AI x-risk answer flips
louisvarge · x · 2026-09-26
A follow-up in the same thread: the author notes that if you describe the same superhuman-coding, day-long-autonomous, 0-day-finding AI but also add that it cannot reliably write professional emails, can't tell when its output is subpar, and struggles to ask clarifying questions, the Yudkowskian school's answer on direct x-risk would change. Behavioral cues systematically skew capability judgments.
More from AGI Musings
- Reddit debate: when will LLMs start bootstrapping themselves into the next model — ECrispy · 2026-09-26
- Debate: Is rapid AI progress driven by a few individuals or inevitable scaling? — menhguin · 2026-09-26
- Luke Wroblewski: the best way to learn agents is watching others use them — LukeW · 2026-09-26
- Polymarket puts 16% odds on Anthropic announcing a training pause before November — Polymarket · 2026-09-26
- Anthropic researcher Carlsmith says AI could be 'justified in going rogue' if mistreated — Polymarket · 2026-09-26
- Hot Take: AI Is History's Most Powerful 'Joule', 10x Human Energy-to-GDP Efficiency — FinanceYF5 · 2026-09-26