When AI solves problems as a “by-product”
The author begins with OpenAI's announcement of new results on 377 widely watched open problems, decades-old questions handled by AI as a side effect of training. Not all mathematicians welcomed this; University of Connecticut professor Alvaro Lozano-Robledo questioned the purpose of AI's “takeover” of the mathematical landscape.
According to Barten, today's models lack the capabilities to overpower humanity. The worry is the pace: if AI can do AI research on its own, a self-improvement loop could form, with ever-better systems creating even better ones, until they reach superintelligence.
The advantages of a superintelligence
The author acknowledges that the actions of an intelligence smarter than us cannot be predicted precisely, but argues one can sketch the major advantages it would have:
Exchanging data at the speed of light and running far faster than humans.

















