Small LLMs Still Bet Up to 82% on an 11th Coin Flip After Ten Heads
This story is from 2026-11-16. It is preserved in the archive; the latest stories are on the live feed.
A study of eight base and instruction-tuned open models found that even when told a coin is fair and flips are independent, their streak bias drops by only a few points, with some betting up to 82% on heads after ten heads in a row.
Read the full story at Towards Data Science ↗
Timeline · 1 report
- 2026-11-16 16:30 · Towards Data Science
After 10 Heads in a Row, Small LLMs Bet Up to 82% on an 11th