[hanasu]

Theoretically the best way to maximize an edge is to just wager larger amounts at all times. This has a beneficial side effect of deviating from 'standard' human play which can be further exploited, although I doubt this the intention of the AI.

[xapata]

Maximum expected value is not always optimal. Bets are an investment with uncertain returns. In normal circumstances, portfolio theory helps us optimize our investments. This reduced-variance format removed the budget constraint and made portfolio theory irrelevant. Overbets made sense to the bot who had only trained in this particular format, but not to the humans who have trained for years with a budget.