Look, it was the most extreme contrast that I could think of, to test out Meta's new 'text-to-audio' AI tech, OK? You can hear what happened when I asked for 'Barbie Girl in the Style of Slipknot' in a bit ...
First, though, let's marvel once more at the weirdness that is 'Zuckerberg as open-source hero', enabling mere mortals such as we to play around with some of the most cutting-edge tech there is. It's almost certainly Yann LeCun , rather than the Zuckster, who's driving this - and it's probably written into his contract that he can do whatever he likes, as one of the Godfathers of AI, as long as he doesn't jump ship from Meta - but even still, never say crazy can't get crazier.
Which brings us to this piece of tech that sounds (in print) pretty amazing - what is sounds like (in sound) isn't necessarily all that, though. I tested it through the very helpful #wiringbits. You can build a server to host it yourself if you're that way inclined (save your time, please) - or just click on this link and describe the sound you'd like to hear.
I think the idea is that, in future, you'll be able to build corporate videos with soundtracks that you've created, including footage of specific incidents that may not have happened but you want to have audio in the background. At present, though, I think that this is one of those "Guys! We've built this and it's amazing!! Now, what shall we do with it??" pieces of tech that is a solution looking for a problem.
So don't take too much time on it yet - it'll improve over time and, who knows, might find that problem it's required to solve after all. In the meantime, if you wanted to hear what it thinks ten seconds of 'Barbie Girl in the style of Slipknot' sounds like, then listen away ....
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comments