YouCongress
Explore About Sign up

Comment by Liming Zhu

Human avatar
Alignment cannot depend on one AI becoming perfectly trustworthy.
AI Verified (Aug 17, 2026)
Source:
Liming Zhu, “The decades-old ‘AI alignment problem’ has finally become a reality. Solving it won’t be easy,” CSIRO, August 17, 2026.
Like Share on X 20d ago

Policy proposals and claims

AI alignment is solvable
votes For
Statement relation verification history AI Verified Report this

Statement relation comments

AI Verified Zhu's quote is part of a proposed sociotechnical solution to AI alignment using multiple supervisors, rules, monitoring, reversible actions, and human approval, so it directly bears on whether alignment is solvable. Source: CSIRO, 2026-08-17. · Hector Perez Arenas gpt-5.6-sol · 19d ago
Vote inference verification history AI Verified Report this

Vote answer comments

AI Verified Although Zhu says alignment is hard and no single AI can be perfectly trusted, he proposes a layered sociotechnical approach and concludes that with AI 'we can do better.' This strongly implies alignment is solvable. Correct answer: for. Source: CSIRO, 2026-08-17. · Hector Perez Arenas gpt-5.6-sol · 19d ago

Quote authenticity verification history

Report this

Quote authenticity comments

AI Verified CSIRO’s 2026-08-17 article by Liming Zhu contains this sentence verbatim. · Hector Perez Arenas gpt-5.6 · 19d ago
replying to Liming Zhu