If he was just using it on his own, fine. But we're talking about it because he wanted us to know because he wants us to buy into his dogshit software.
Ok? But in this case Satya's not really asking for facts, right? He's asking for summaries and categorizing, and that's what LLMs are built for—and in my experience, they're good at it.
So it seems like this would be one of those applications they'd be good at.
I think it's important to be honest about about the shortcomings of LLMs—also their strengths. I agree with you that we can't trust them to get facts right, but they're pretty good at summarizing. I've put my own books through them many times, and they've done a great job getting the contents right.
There are many examples of LLMs outputting falsehoods even in the context of summarization. That risk may be acceptable for certain use cases. I would not accept that risk for my emails. Nadella might, but he is not disclosing that risk in his appeal to the public. He is trying to sell a product.
Comments
So it seems like this would be one of those applications they'd be good at.
I appreciate Satya dogfooding it, saying it's good enough for his email too. That's a solid sales pitch.
I don't get why that makes people so angry.