"James Watt's steam engine is just an expansion of something called the Aeolipile, the Ancient Greeks had it, yawn, how boring, why would anyone think it's significant."
"James Watt's steam engine is just an expansion of something called the Aeolipile, the Ancient Greeks had it, yawn, how boring, why would anyone think it's significant."
The steam engine does useful work. Bit of a cornerstone difference there. This isn't the steam engine. It *might* be the Aeolipile.
People are absolutely doing useful work with modern LLMs that they were unable to do with the Markov chain or gradient descent demo apps you or I wrote in a undergrad CS course. The datasets are part of that, but that doesn't invalidate the point.
I dispute both "Useful" and "Work", together and independently. People are generating text that wasn't worth writing and slop code that is going to create a tech debt apocalypse in fairly short order.
It's easy to say when you speak English. I use DeepL all the time for translations. This one is done on it too. Making automatic translations is obviously very useful. So are subtitles.
Translations are of course the least hyped end of the spectrum when it comes to AI, but it allows you to follow news in different languages. For example in French, Russian and German.
The chances that you convince him that you find it useful are zero because you can't reason someone out of a position they didn't reason themselves into.
(One of the things that makes Watt's steam engine meaningful and revolutionary compared to the Aeolipile was that it was adapted to do useful work by being able to scale to real-world loads. This...sounds familiar.)