Back to feed
OpenAI Blog·

Deep double descent

Signal
75
Hype
25
In three linesOpenAI demonstrates that double descent phenomenon occurs in CNNs, ResNets, and transformers: performance improves, then worsens, then improves again with increasing model size, data size, or training time. The effect is universal but not yet fully understood.
Read source
Your take?
OpenAIPapersBenchmarksReasoning

Summary generated by Claude — human-verified