According to The Information.OpenAI Next Generation Flagship Model Orion of progress may not be as great as the predecessor model. Internal testers found that theWhile Orion outperforms existing OpenAI models, the improvement is small compared to the leap from GPT-3 to GPT-4.
This means that the pace of AI progress may be slowing. In fact, Orion may even be inferior to previous models in some areas, such as coding.
To address this challenge, OpenAI has set up a foundational team to look at how to continue to improve model performance as training data dwindles, according to the report.New strategies include using AI models to generate synthetic data to train Orion, as well as additional optimizations after model training.
OpenAI has not yet responded to this report as of IT Home's press release. Previously, the company had responded to reports about plans for its flagship model by saying, "We have no plans to release a model called Orion this year."