Fake Gemini benchmark lists appear fourth year running, complete with bogus 'Astra' scores
ChrisGPT · x · 2026-09-18
ChrisGPT calls out a recurring phenomenon: someone publishes a Gemini benchmark list that looks plausible but is obviously fake — the fourth year in a row, each time containing catastrophic errors like fabricated "Astra" scores. The author muses that people believe it anyway and begs the fake-list maker to at least produce one without glaring flaws.
More from Fun
- Yacine on why he tweets: poking the social fabric and predicting reactions — yacineMTB · 2026-09-20
- Yacine says Chinese group chats sharing Anthropic interview questions are real — and hilarious — yacineMTB · 2026-09-20
- Meme: How it feels not using Jev, Instinct or Wispr Flow in this economy — wateriscoding · 2026-09-20
- A Heuristic for Spotting AI Replies: Could It Be Its Own LinkedIn Post? — jh3yy · 2026-09-20
- Dario and Sam said we must slow down last week — so do they believe it? — _TheWolfOfWalmart_ · 2026-09-20
- Yacine tells followers: my tweets are inside jokes, don't take them at face value — yacineMTB · 2026-09-20