Scaling laws, scaling... with compounding efficiency!
On held-out code, we matched DeepSeek-V4-Pro-Base's pre-trained score using 12x less training compute.
Microsoft matches DeepSeek-V4-Pro-Base code score with 12x less compute
AISummary
Mustafa Suleyman says Microsoft matched DeepSeek-V4-Pro-Base's pre-trained score on held-out code using 12x less training compute. He attributes the result to compounding efficiency gains.
Post on XView on X
Source: Mustafa Suleyman · x.comPublished · added here