The post doesn't deny that the utility function is concave, but it says it can't reasonable be that much concave as needed to explain rejecting +55/-50 even odds bet with 15,000 on the gambler's account.

What I'm saying is that you've grasped how risk aversion is traditionally modelled: though the gain in terms of units is the same as the loss, the consequences of losing are much worse than the consequences of winning (ie being homeless versus owning a second home).

My point is that if you use that model to explain why people turn down small bets (50-50 on winning $55, losing $50 when the people are reasonably well off), then the model predicts stupidly risk aversive behaviour for larger bets, that don't correspond to what people do in practice.