It is genuinely frustrating to participate in LessWrong as a community that effectively gates itself behind AI vetting tools. The community claims to critically engage with the evolving AI ecosystem, yet moderators rely on unvetted LLMs and AI tools to automatically tag and assess posts. Currently, it seems that any...
Per BlueDot Impact's Technical AI Safety course, we should operate on the assumption that our primary defenses will eventually fail. Models will learn to deceive evaluations, dangerous capabilities will emerge unexpectedly, and edge cases will slip through. This "assume harm" mindset is highly relevant to current industry dynamics, particularly the...
A diagnostic look at how AI scribes, the most widely deployed clinical AI in the world right now, are slipping into permanent medical records under governance frameworks designed for office software. In May 2026, the Office of the Auditor General of Ontario published a special report on AI use across...