Will OpenAI’s hyped Orion disappoint?

A new report from The Information suggests that OpenAI’s next flagship model may not be as leap forward as its predecessors. Previously, we talked about the next and famous model of OpenAI; A model known by the code name Orion based on Sam Altman’s tweets on the X social network; However, the report from employees who tested Orion doesn’t sound so encouraging; The staff found that, although it outperformed existing OpenAI models, it provided less improvement than jumping from GPT-3 to GPT-4.

In other words, the rate of recovery seems to be slowing down; In fact, the Orion may not be reliably better than previous models in some areas, such as coding.

In response to The Information’s report, OpenAI commissioned a team to figure out how the company could continue to improve its models; These new strategies reportedly include training Orion on synthetic data generated by AI models.

OpenAI did not immediately respond to a request for comment; The company has previously said it has no plans to launch a model codenamed Orion this year, despite evidence of Orion development.