Hacker News Reader

If turn the temperature down for tests, they won’t match production behaviors. If you turn it down too far in production, the output will just be bad.

View on HN

Isn't the goal of the author to get reproducible behavior out of the agent though? I would thinking turning the temperature down would serve that production goal too.

right, this is the whole problem - the stochastic behavior is both the goal and the problem. If you want your tests to match production, you need to get a reasonable sample size, which costs real money.

2 things.

1. If you turn the temperature down too far, the output is just bad and no amount of running prompts optimization will let you hill climb your way to good performance.

2. It’s not about determinism vs non-determinism. It’s about chaos. A perfectly deterministic model is still chaotic. Meaning that very small changes to the input result in very large changes to the output.

Turning temperature down doesn’t actually get you predictable or reproducible behavior across different inputs.

Thank you, I forgot about the chaotic aspect, too - that’s the other part that makes this brutal. The bot performs perfectly on your tests, but your customer abhors the Oxford comma, so you, your marketing team and your test suite can go to hell.

Reply on news.ycombinator.com