News for hackers. Stuff that matters.
Re: by AlbertoRomGar on Tuesday April 12, 2022 @08:11PM (#31009726)
DeepMind just published a new paper that dismantles the trend of building ever-larger language models, established by Kaplan and colleagues in 2020.

They found another factor that matters as much as model size: Data. The best models today are undertrained and oversized!

There are no comments on this comment yet.