Astra 6.2 keeps setting its puzzle games in the same odd 'Bellwether Station' location
ZeroStateReflex · x · 2026-11-19
Ethan Mollick asked Astra 6.2 to build a puzzle spy game with any setting it wanted — it chose "Bellwether Station." Others then found this is a deep, recurring pattern for the model, a weird attractor presumably inherited from training data. A fun window into hidden training traces.
Related event: GPT-6.2 Keeps Setting Games in 'Bellwether Station'(2 posts)→
More from Fun
- Demis Hassabis fuels Gemini 4 family speculation by sharing Breaking Bad-style meme — demishassabis · 2026-11-19
- Matic robot vacuum started up while cleaners were over: 'felt like I got caught cheating' — yungcontent · 2026-11-19
- David Deutsch: Musk's Childhood Trance Shows Why No Behavioural Test Can Detect AGI — Paimaamu · 2026-11-19
- Heartbreaking: someone got paid way more than you to publish the same findings — minhnguyen · 2026-11-19
- 49:1 male-to-female ratio at AI summit, where even coffee lines debate inference costs — TarynSouthern · 2026-11-19
- First look: Ike Barinholtz plays Elon Musk in Luca Guadagnino's 'Artificial' about OpenAI's Altman saga — Polymarket · 2026-11-19