OpenAI codenamed the latest pre-training model of Doug

Recent news suggests that OpenAI is moving forward with a new, large-scale model, code-named Doug, which will be the largest pre-training project to date and is expected to be available no later than November this year. Since the publication of GPT-4o in May 2024, OpenAi has not introduced a large-scale updated backbone model, which has resulted in a leap forward in model capacity, mainly through training, but industry has indicated that this is a long-term problem of marginal gain erosion. In the face of competitive pressure, OpenAI restarted and adjusted the bottom-seat pre-training strategy, which had been validated by the Garlic model as a major extension of these improvements. The industry believes that if Doug goes well, it means that OpenAI's return to a large base pre-training track will open up a new round of model competitions that will strengthen the learning system in line with its maturity, and that the future AI performance ceilings will have greater imagination。