OpenAI's next AI model won't be a big leap forward
What's the story
OpenAI is said to be working on its next flagship model, codenamed Orion. However, as The Information reports, the new model may not be a major leap over its predecessors. Employees who have tested Orion have observed that while it is better than existing models from OpenAI, the pace of improvement is slower than what we saw in the transition from GPT-3 to GPT-4.
Performance concerns
Orion's performance in certain areas may not surpass predecessors
The report further suggests that Orion's performance may not always exceed previous models in certain areas, like coding. This points to a possible slowdown in the pace of improvement for OpenAI's models. The company has not yet responded to requests for comment on these findings. Previously, OpenAI said it had no plans to release a model code-named Orion this year.
Strategic response
OpenAI forms team to address AI improvement slowdown
In light of the apparent slowdown in AI improvement, OpenAI has reportedly established a foundations team. The team is responsible for investigating how the company can continue to improve its models even as the supply of new training data dwindles. The strategies being explored include training Orion on synthetic data produced by AI models and making additional improvements during the post-training phase.