Ex-Anthropic staffer urges slowing AI capability gains amid uncertainty

Former Anthropic staffer Flomerboy cited cases like Claude scoring worse with a zoom tool due to pixel-by-pixel viewing, arguing that racing ahead amid such uncertainty without changes is irrational and safety should be strengthened.

2026-09-13 ~ 2026-09-13 · 3 related posts

• Adding a zoom tool made Claude worse: it was inspecting every pixel via code — Flomerboy · 2026-09-13
• Anthropic alum: unexpected AI problem-solving plus real misalignment cases make the case to slow down — Flomerboy · 2026-09-13
• Under deep uncertainty, charging ahead unchanged makes no sense, argues Flomerboy — Flomerboy · 2026-09-13