> But the problem type of "summarize a biased source and try to remove the bias" isn't among the ones I've seen LLMs being tested for, and this is what Elon's project lives and dies by.

And if you believe that you’ll believe anything. “Try to _change_ the bias” would be closer.