I agree it's better, but it's still awful at it. But I think it's one of the low hanging fruits to improve, though, just by reinforcing the limits of what it knows.
With respect to the competing draws here, I'm not sure they necessarily compete that much. E.g. being able to ask it to speculate but explain justifications or being able to provide a best guess or being able to ask it to just be as creative as it wants, would be sufficient. Alternatively one that knows how to embed indication of what it knows to be true vs. what is speculation or pure fiction. Of course we can "just" also train different models with different levels of / types of reinforced limits. Having both a "anything goes" model and a constrained version that knows what it does know might both be useful for different things.
With respect to the competing draws here, I'm not sure they necessarily compete that much. E.g. being able to ask it to speculate but explain justifications or being able to provide a best guess or being able to ask it to just be as creative as it wants, would be sufficient. Alternatively one that knows how to embed indication of what it knows to be true vs. what is speculation or pure fiction. Of course we can "just" also train different models with different levels of / types of reinforced limits. Having both a "anything goes" model and a constrained version that knows what it does know might both be useful for different things.