It is generally true with NNs that they often cannot admit they don't know something and will output some ridiculous prediction anyway, but I feel it is especia

It is generally true with NNs that they often cannot admit they don't know something and will output some ridiculous prediction anyway, but I feel it is especially bad with language models, and ESPECIALLY bad with something like this "galactica". Forget that it's pretty bad in its predictions, even if it was not the case, you just cannot rely on it at all, when it can so nonchalantly produce some plausible bullshit. How can you even distinguish? This even sounds like something Lem would totally write. Except he just didn't. (And Polish "translation" isn't right, but I guess you couldn't tell if you don't speak any slavic language. And even then you might believe that "kotel" is "a cat" in Polish — who knows, plenty of weird discrepancies between languages!)