Comment by woah
3 days ago
> the underlying tech has plateaued, but now they're brute force trading extra time and compute for better results
You could say the exact same thing about the original GPT. Brute forcing has gotten us pretty far.
3 days ago
> the underlying tech has plateaued, but now they're brute force trading extra time and compute for better results
You could say the exact same thing about the original GPT. Brute forcing has gotten us pretty far.
How much farther can it take us? Apparently they've started scaling out rather than up. When does the compute become too cost prohibitive?
Until recently, training-time compute was the dominant cost, so we're really just getting started down the test-time scaling road.