Comment by dragonwriter

> That said, it's concerning to see the reported probability for getting a 4 on a die roll is 65%.

Finding that an LLM is biased toward inventing die rolls that are the median result rounded to an available result by the most common rounding method is...not particularly surprising. If you want a fair RNG, use an RNG deigned to be fair, not an LLM where that would be, at best, an emergent accidental property.