Extreme Harness Engineering for Token Billionaires: 1M LOC, 1B toks/day, 0% human code, 0% human review — Ryan Lopopolo, OpenAI Frontier & Symphony
A Latent Space deep dive reveals OpenAI's 'Dark Factory' testing infrastructure—over 1M lines of code and 1B tokens/day, with automated evals replacing human review for frontier model validation.
Excerpt
We shed light on OpenAI's first Dark Factory for the first time.
Read at source: https://www.latent.space/p/harness-eng