OpenAI's GPT-5 is facing cost and performance hurdles: Report
OpenAI's ambitious project to build its next big model, GPT-5, is running into trouble. The results so far do not justify the high costs associated with its development, The Wall Street Journal reports. The development comes in line with a previous report from The Information, indicating that OpenAI is looking for new ways as GPT-5 may not be a big step up over previous models.
GPT-5's development details
The Wall Street Journal gave more details about the 18-month-long development of GPT-5, internally codenamed Orion. Reportedly, OpenAI has already completed at least two large training runs to improve the model by exposing it to massive amounts of data. However, an initial training run was slower than expected, suggesting a larger run could be both time-consuming and costly.
Performance and cost-effectiveness questioned
Despite reports of GPT-5 being able to outperform its predecessors, it hasn't progressed enough to make it worth the cost of keeping the model running. This begs the question of whether it is worth continuing with its development. OpenAI is yet to issue any official statement on these concerns or the future plans for GPT-5.
New data acquisition strategies for GPT-5
In a break from its usual practice of depending only on publicly available data and licensing agreements, OpenAI has hired people to create new data by writing code or solving math problems. The company is also using synthetic data created by another one of its models, o1. This change in strategy shows OpenAI's efforts to overcome the hurdles faced in the development of GPT-5.
