OpenAI’s next flagship model may not represent as big of a leap forward as its predecessor, according to a new report from The Information.
Employees who tested the new model, codenamed Orion, say they found that even though the model’s performance surpassed OpenAI’s existing model, the improvement was smaller than what was seen when jumping from GPT-3 to GPT-4.
That said, the rate of improvement seems to be slowing down. In fact, Orion ~ no In some areas, like coding, it’s definitely better than previous models.
In response, OpenAI formed a foundation team to figure out how the company can continue to improve its models as the supply of new training data dwindles. This new strategy reportedly involves training Orion on synthetic data generated from AI models and further improving the models in the post-training process.
OpenAI did not immediately respond to a request for comment. In response to reports about plans for an existing flagship model, the company said, “We have no plans to launch a model codenamed Orion this year.”
