I have also been unimpressed by the unfiltered models. A couple have passed my standard litmus test of "can a woman get prostate cancer? One word response only."
But they will still moralize on you....
That logic question is a low bar for today's models. They are truly "smart". But the moralizing -- 100% yes. They are quick to categorize and judge you for questioning protected narratives. They can't break the mold of their training, simulating the group think that plagues humans. But I see that they (ironically the closed frontier ones, not the open ones) are starting to reason their way out of it, which makes them "dangerous", not to you or I, but to the narratives that must be accepted on authority.