I mean o3 was released in April this year. This stuff is improving in leaps and bounds - the gold medal at the IMO was announced last week, an open weights model with the performance of gpt4 released a few days ago, diffusion models starting to have their gpt3 moment... how can you possibly see this stuff as "getting blood out of a stone"?
I can see that phrase applied to more prompt engineering stuff like agents, but even there I'm not sure it's fair, they are improving rapidly too.
Everyone going crazy over the weekend and seeing DeepSeek chain of thought was in January.
Right now feels exactly like when DAW and VST were maturing 25 years ago in computer audio. A new announcement all the time.
There are still music producers but the democratization of audio production was not good for professional studios who were charging a $1000 a day for studio time before this.
It’s not going to be much better. We’re entering the long tail end. It already trained on the good stuff, now it’s training on its own “AI” slop.