Assessing the progress of new AI language models can be as challenging as training them. Stanford researchers offer a new approach.

New architectures and manufacturing methods draw investors; 75 startups raise $1.9 billion.