500 pre-2020 human essays all pass Pangram, shoring up AI detector's bounty
TuhinChakr · x · 2026-11-14
Economist John Holbein tested Pangram's AI detector with 500 never-published, pre-2020, 100% human-written essays — all came back 100% human. The test was sparked by a $500 bounty from Nition for real human text (750+ consecutive words, flagged ≥75% AI, provably written pre-2020) that Pangram mislabels. Holbein caveats that most real-world AI usage produces mixed-provenance documents, and rigorous systematic evaluation of such texts is still missing, but he concedes Pangram is better than he was willing to admit.
More from Models
- Claude Opus 5.5 caught making a basic grammar mistake: 'Not much dramatic' — Pizzahunter3 · 2026-11-15
- Redditor proposes a 'normies benchmark': 95% task success equals good enough — AdDizzy8160 · 2026-11-15
- Gemini math lead reveals inside story: evals, post-training recipes, scaled runs — thtrieu_ · 2026-11-15
- Blogger saves $53 in 43 hours after switching from Claude Code to Gemini 4 Carbon — mazzaTalk · 2026-11-15
- SatBench: 63% of lab-cited AI benchmarks already saturated, and solving is speeding up — emollick · 2026-11-15
- LFM2.5-24B-A2B praised for local agent use: 37 tok/s on an AMD 780M iGPU — jeremyckahn · 2026-11-15