Bill Gates warns against unrealistic expectations for GPT-5 performance

In 2023, Bill Gates expressed skepticism about the advancements of OpenAI's upcoming AI model, GPT-5. During an interview with the German newspaper Handelsblatt, he stated that while many at OpenAI expected significant improvements, he believed that generative AI had reached a plateau. Gates highlighted the exceptional advancements between GPT-2 to GPT-4 but cautioned against inflated expectations. With the launch of GPT-5, user feedback revealed numerous complaints about performance issues, such as bugs and unremarkable responses, causing OpenAI to revert access to GPT-4 for some users. CEO Sam Altman acknowledged the shortcomings and promised fixes, emphasizing future models would achieve considerable improvements. Despite the mixed performance results — for instance, GPT-5 scored 56.7% on the public SimpleBench leaderboard, ranking it fifth among tested models — OpenAI maintained it was the smartest system yet. On SWE-bench Verified, the model achieved a more notable 74.9%. Gates’ predictions were validated as the constraints he mentioned regarding the limits of generative AI became apparent upon GPT-5's release, leading to backlash from users and analysts alike. Additionally, he discussed the need for generative AI to become cheaper, more reliable, and practically useful in everyday applications, such as providing trustworthy health advice via mobile devices. Given the expensive nature of compute and semiconductors, Gates urged the need for new research pathways for developing better-performing AI. The hype surrounding GPT-5 created unrealistic expectations for its performance. OpenAI's leadership, including CEO Sam Altman, spoke highly of the model, suggesting it could eventually surpass human intelligence in its capabilities. However, this overpromising contrasted sharply with the early feedback. Altman noted that productivity and UX improvements, like better clarity on model responses, would soon be implemented. Meanwhile, Gates maintained a pragmatic outlook, emphasizing that genuine progress in AI applications would stem not just from scaling existing models but from innovative research focus to overcome current limitations. Addressing ongoing AI challenges remains critical, especially in the face of rapid technological development and public scrutiny regarding AI capabilities in various sectors, including healthcare and programming.