OpenAI’s Orion model faces scaling challenges

The news: OpenAI’s upcoming model, codenamed Orion, won’t improve on its predecessor as much as the company’s previous versions have, owing in part to a weakening pool of training data for generative AI (genAI) models.

  • Orion outperforms GPT-4 with language tasks, per The Information, but isn’t reliably better with coding or math work.
  • The next-gen model is expected to launch this year and will likely be available only to partnering companies at first.