Reports highlight challenges in advancing AI model code-named Orion.
OpenAI’s ambitious development of its next-generation AI model, GPT-5, code-named Orion, is reportedly running behind schedule, with performance improvements that fall short of justifying its high costs, according to a new report in The Wall Street Journal (WSJ).
The 18-month-long project has encountered delays during large-scale training runs, which are integral to improving the model’s capabilities. Sources told the WSJ that an initial training run took longer than expected, raising concerns about the time and expense required for a larger run. While GPT-5 has demonstrated improvements over its predecessors, its current advancements are reportedly not substantial enough to offset the costs of its development and deployment.
An earlier report by The Information suggested OpenAI might be rethinking its approach to GPT-5, as it may not offer as significant a leap forward as previous iterations like GPT-4. The WSJ adds that OpenAI has taken innovative steps to refine the model, including creating custom datasets.
In addition to using publicly available data and licensing agreements, OpenAI has hired individuals to generate new data, such as writing code or solving complex math problems. It is also leveraging synthetic data produced by another AI model within its ecosystem, known as o1.
The delays and challenges highlight the increasing complexity and cost of advancing AI systems as they approach the limits of current technologies. OpenAI has yet to comment on the recent reports, though the company previously stated it would not release a model code-named Orion this year.
Comments are closed.