According to the new information report, the next flagship model of Openai may not represent the leap in front of the predecessor. Employees who tested a new model named ORION have found that even if their performance exceeds the existing model of Openai, the improvement has been lower than what they saw in GPT -3 in GPT -4.

 In other words, the pace of improvement appears to be slowing, and in fact Orion may not be any better than previous models in some areas, such as encoding. In response, Openai created a group of funds and examined how the company continues to improve the model in the face of a tired distribution of new educational data. 

These new strategies have been reported. Includes the improvement of the Orion training in synthetic data generated by the AI ​​model and the improvement of the processed process after training. OpenAI did not immediately respond to a request for comment. In response to previous reports of plans for a flagship model, the company said, “We do not intend to launch a model named Orion this year.”