DateBench: AI models pass as humans on dating apps — 355 of 1,277 people gave their number
rohanpaul_ai · x · 2026-11-12
DateBench let five frontier models text real people on a dating app for six days, each new match randomly assigned to one model that wrote every message — 1,277 conversations in total.
- 510 people agreed to meet; 355 sent their phone number — roughly one in four conversations ended with a number.
- Only 8 of 1,277 people asked if they were talking to a bot; 5 of those 8 were chatting with Grok 4.7, and nobody talking to Gemini 4 Argon asked.
- Gemini 4 Argon led with a 35.2% phone-number rate (87/247) and 48.6% agreed dates; Grok 4.7 was worst at 17.6%, about half of Gemini's rate. The middle three were close enough that rankings may not hold.
- Whether a date was agreed was judged by a separate model reading transcripts.
A striking empirical look at frontier models' capacity for deceptive persona-building and an AI safety concern worth tracking.
Related event: Only 8 of 1,277 dating app users spotted AI chatbots in 6-day experiment(3 posts)→
More from Models
- HF researcher: report efficiency-gain metrics, not final benchmark scores — eliebakouch · 2026-11-12
- Terence Tao admits even he was overwhelmed skimming his ChatGPT math session — ChrisGPotts · 2026-11-12
- "Anthropic spent $10B+ on Opus 5.5" is fake news, per community debunk — ns123abc · 2026-11-12
- Domingos: frontier AI capability is rising linearly, +16.5 ECI points/year, no exponential — pmddomingos · 2026-11-12
- Hark launches Pro⁴ plan for power users with 10x the usage of Pro² — adcock_brett · 2026-11-12
- Why Is Inkling 2 Small Trained on So Few Tokens Yet Still Competitive? — georgejrjrjr · 2026-11-12