Microsoft's research team has announced that they have succeeded in drastically reducing the computational cost of large-scale language models by setting the model weights to only three values: ``-1'' ...