Ask HN: Isn't most of the magic of AI the embeddings?

AI allowing the pairing of 'ideas' and appropriate tools without the need to dive into surrouding details for task at hand. aka telepresence/tricorers/ etc. [0][1][2][3][4]

[0] : https://www.tricorderproject.org/

[1] : https://www.youtube.com/watch?v=4Uwbq9e9Ftw

[2] : https://www.unit-led.com/mandalorian-led-wall

[3] : with 3d printer, https://www.youtube.com/watch?v=aRIVSDzVzlE

[4] : https://www.startrek.com/news/10-cutting-edge-real-world-sta...

That’s like saying the magic of a model is its predictions. The magic is really the process of gathering data, cleaning it, creating a model architecture, and meaningful objective function for the model that creates the embeddings.

Embeddings themselves aren’t new. It’s how they’re being trained - and the incredible scale - that’s new.

I feel the need to write yes 50 times.