The real unlock with LLMs isn’t doing old things faster. It’s doing things that were not possible before. We’re seeing tools that let you: – Hum a melody and get a finished track (Bovo Music) – Produce Hollywood-level scenes, no training or crew (Intangible) – Spec product features like a seasoned lead, right out of school (ProductNow) This isn’t automation. It's a capability shift. The gap isn’t between fast and slow; it’s between can and can’t. What are you doing with AI that was impossible 18 months ago?
True, though the research on topics like "melody recognition" is done since long ago (I remember doing undergrad project on this back in 2004! So this was present definitely long before that), it certainly was a different field back then wrt techniques due to lack of data and compute. And of course, many things were just not possible/good enough just because of that and the techniques that were present due to restrictions. That's the reason suddenly those things feel achievable now with restrictions getting lower and techniques evolving because of that!
Very nice. And Suno of course - make music where you previously couldn't.
Felix Wong That "can vs. can't" distinction is the perfect way to frame it. I think the next big shift is seeing these new capabilities move beyond individual tools and into the hands of entire teams. When non-technical people can build their own custom workflows with this power, that changes how a business fundamentally operates.
Grateful to know some of the people building this next wave: Donald Szeto, Charles Migos, Bharat Vasan, Tript Singh Lamba. You're pushing the edge.