A team of researchers proved a simple mathematical law — the “linearity theorem” — that lets engineers predict, in advance and with precision, exactly how compressing an AI model will affect its intelligence. No more guess-and-check. No more crossing your fingers and hoping a 70-billion-parameter model doesn’t collapse into gibberish after compression.

A global research team just proved a simple mathematical law that lets engineers shrink massive AI models onto ordinary laptops and phones — without guessing, and without dumbing them down.