A more worthwhile aspiration would be narrower and more concrete, such as building AI systems that reduce medical diagnostic errors by 30%. Or in education, improving math proficiency in students by 15%. Or systems that could enhance energy grids to reduce carbon emissions by 20%. Such goals not only have clear metrics for success but serve concrete human needs, just as AI builders like Altman and Hassabis originally envisioned.
It's not one or the other. AI labs are building foundational models and the next layer will build on top of them to improve medical diagnostics, education, etc. There are plenty of LLM wrapper startups with those exact goals.
Trouble with many of those small number goals is that they don't really get you anywhere.
MYCIN was an accurate medical diagnosis and prescription system in the 1970s [1] but was rejected by the world. It's not clear that test scores are part of the solution or part of the problem in education. It's not enough to reduce carbon emissions by 20%, we have to reduce them by 100% and the lack of a plan to do so makes anything a political loser: people who are opposed to taking action are going to be opposed anyway, people who want to see action are going be despirited, the ecosystem doesn't care but it will dump our asses off Spaceship Earth anyway. (And it just takes a 20% increase in energy demand to wipe out that 20% efficiency gain)
