Unit economics vs. the singularity by Thomaaas

This is a con­densed ver­sion of the es­say linked above.Not long ago, it was fash­ion­able to ask “how would the AI labs ever turn a profit?” In the­ory, ever-in­creas­ing train­ing costs would swal­low all rev­enue, as labs com­pete to stay at the fron­tier. If a lab stopped train­ing bet­ter mod­els, their com­peti­tors would catch up, un­der­cut them and take their mar­ket share. A lose-lose.An­thropic is poised to prove a lab can be prof­itable. How­ever, if progress slows, low switch­ing c...