o1 sudden insider exposure? Google reveals principles earlier, big models don't have moats with software alone

Less than 1 week after its release, the moat of OpenAI's strongest model, o1, is gone. It has been discovered that a Google DeepMind paper published in August reveals that the principles are almost identical to the way o1 works. This study showed that increasing test-time computation is more effective than expanding model parameters. Based on the computationally-optimal (compute-optimal) test-time computation scaling strategy proposed in the paper, a smaller base model can outperform a 14 times larger model on some tasks. Some users said: this is almost the principle of o1. (Quantum Bit)