Recently, OpenAI Chief Technology Officer Mira Murati clarified rumors about the company’s internal AI system capabilities. There has been speculation that OpenAI has an advanced AI system that far exceeds the public model. Murati said that there is no significant difference between the capabilities of the AI models currently in the laboratory and those available to the public. This statement triggered discussions in the industry about the bottlenecks in the development of large-scale language models, and was also inconsistent with some people’s expectations for the future speed of AI development.
OpenAI CTO Mira Murati clarified speculation about the highly developed AI systems within the company: In an interview, she made it clear that the AI models in the OpenAI research lab are currently no more capable than publicly available models.
"We have these powerful models in the lab, and they're not a huge improvement over what's freely available to the public," she told Fortune.
Her statement also implies that if the current internal system is only slightly better than GPT-4, the performance of OpenAI's public AI models is unlikely to improve significantly in the near future.
This also means that significant improvements in the performance of OpenAI’s public models are unlikely to occur in the near future. After the model is trained, it requires months of fine-tuning and security testing before the model can be released to the market.
Critics believe that this also confirms that the expansion of this type of AI technology has reached a bottleneck. In fact, since the release of GPT-4, there has been no significant progress in the basic capabilities of large language models, especially in logic.
Microsoft founder Bill Gates also predicts that compared with the huge progress from GPT-2 to GPT-4, the leap from GPT-4 to GPT-5 will be smaller.
However, OpenAI CEO Sam Altman and Microsoft CTO Kevin Scott continue to promise significant progress, which Scott says will happen this year.
Murati’s statement provides a new perspective on the current status of AI development, and also brings new thinking space for the future progress of large-scale language models. Although breakthroughs are limited in the short term, in the long run, the continued development of AI technology is still worth looking forward to.