I'm guessing this is a side effect of mistakes in the reinforcement learning face. It'd be really easy to build a reward model that favors passing tests, withou

I'm guessing this is a side effect of mistakes in the reinforcement learning face. It'd be really easy to build a reward model that favors passing tests, without properly measuring the quality of those tests.