According to The Information, OpenAI’s latest AI model, Orion, is facing challenges, delivering smaller performance improvements compared to its predecessors. While the model achieved GPT-4-level performance after just 20% of its training, experts within the company report that Orion is not consistently outperforming GPT-4, especially in tasks like coding.
Despite Orion’s strengths in language tasks, researchers suggest that the model may not provide the same groundbreaking advancements that previous versions, like GPT-3 and GPT-4, did.
According to an OpenAI employee:
“Some researchers at the company believe Orion isn’t reliably better than its predecessor in handling certain tasks. Orion performs better at language tasks but may not outperform previous models at tasks such as coding.”
This issue comes at a critical time for OpenAI, following a recent funding round where the company raised $6.6 billion. With this influx of capital comes increased pressure from investors to deliver meaningful progress.
However, the diminishing returns from AI model improvements could threaten future fundraising opportunities if expectations aren’t met.
A significant contributing factor to Orion’s limited improvements is the decreasing availability of high-quality training data. AI companies, including OpenAI, face a looming shortage of publicly available human-generated text data, which is expected to be depleted between 2026 and 2032.
To tackle the challenge, OpenAI is rethinking its development strategy, shifting focus to enhancing models after their initial training, which could lead to new scaling laws.
Despite these hurdles, OpenAI continues to innovate. The company is navigating a delicate balance between maintaining investor confidence, pushing technical boundaries, and addressing challenges like leadership departures, which may further complicate its AI development.
