Right thesis, wrong structure
Leopold Aschenbrenner (@leopoldasch) published a 165-page essay in June 2024 arguing that AI capability was scaling faster than institutions understood, and that the buildout required to sustain it …
Article · @tadle_com