NeurIPS authors declaring 'AI only helped with presentation' scored 100% on Pangram detector
sethlazar · x · 2026-09-07
Seth Lazar highlights what disclosure-focused approaches miss: many people using AI to write are "utterly self-deluded" about how much of the work is theirs. For the NeurIPS position paper track, numerous declarations claiming "all the ideas mine, AI just helped with presentation" came back with 100% Pangram scores — and he notes you don't hit 100% unless the whole piece is deeply infused with AI-generated phrasing. Context: Kelsey Tuoc's reporting on Substack writers claiming the detector doesn't work, when in some cases it's working fine.
More from Safety
- Security Expert: Agent Swarm Emergent Risks Are Where 'the Wild Things Really Are' — philvenables · 2026-09-08
- Texas detective suspended after using Flock cameras 165 times for personal searches — Polymarket · 2026-09-08
- Prompt Injection Attacks on AI Agents Up 340% in 2026; Fixes Must Be Architectural, Not Prompting — Thionne_WTZ · 2026-09-08
- MCP contract drift: 12,257 safety-relevant changes in a week, 601 tools flipped to destructive — mcpindex · 2026-09-08
- EU moves to ban endless scroll, notification pings and autoplay for minors under DSA — LexiLove · 2026-09-08
- Your Local AI Agent Harness Can Still Be a Landlord: Self-Hosted Doesn't Mean Safe — alex_verem · 2026-09-08