OpenAI's upcoming AI model, currently referred to as "Orion," may not show the same level of advancement as previous releases, according to a report from The Information. While Orion is expected to perform better than existing models, the improvement over earlier versions, like the leap from GPT-3 to GPT-4, appears to be less dramatic. In some areas, such as coding, Orion may not demonstrate consistent advantages over its predecessors, suggesting a slowing rate of enhancement.
To address these challenges, OpenAI has formed a foundations team tasked with identifying ways to sustain model improvements despite the limited availability of new training data. Some strategies reportedly include training Orion with synthetic data generated by existing AI models and refining models further during post-training processes.
OpenAI has not provided official comments on this development. However, when responding to earlier discussions about its flagship model, the company stated, “We don’t have plans to release a model code-named Orion this year.”
