OpenAI’s ‘Orion’ Model Shows Diminishing Returns —Report

Artificial intelligence company OpenAI’s next flagship model, code-named Orion, demonstrates smaller performance improvements compared to previous iterations, The Information has reported, citing company employees familiar with the matter.

Internal testing reportedly showed that while Orion surpasses existing models overall, the advancement gap is less significant than that observed between GPT-3 and GPT-4, the company’s previous major releases. The model struggles to show consistent improvements in coding capabilities over previous versions, according to the sources.

OpenAI has formed a foundations team to tackle challenges related to limited training data availability. The initiative explores new approaches, including the use of AI-generated synthetic data and enhanced post-training optimization techniques, the report said.

The company faces additional concerns about potentially higher operational costs for running Orion in its data centers compared to current models, one source said.

OpenAI previously told TechCrunch that it has no plans to release a model code-named Orion this year.

Information for this story was found via the sources and companies mentioned. The author has no securities or affiliations related to the organizations discussed. Not a recommendation to buy or sell. Always do additional research and consult a professional before purchasing a security. The author holds no licenses.