2 ms·

Reductive. Doesn’t explain Deepseek.

• FergusArgyll 1y ago
  Deepseek story was way overblown. Read the gpt-oss paper, the actual training run is not the only expense. You have multiple experimental training runs as well as failed training runs. + they were behind SOTA even then