Since its groundbreaking launch last year, OpenAI’s ChatGPT has revolutionized the world of artificial intelligence, impressing users with its human-like question-and-answer interactions. The chatbot’s potential to challenge Google’s search engine dominance had sparked excitement among industry insiders. However, recent developments have led to growing concerns over ChatGPT’s declining performance, leading some to question the future of generative AI technology.
Researchers from Stanford and UC Berkeley conducted a comprehensive analysis of different versions of ChatGPT, comparing March 2023 and June 2023 iterations of GPT-3.5 and GPT-4. The study focused on four essential tasks: math problem-solving, answering sensitive questions, code generation, and visual reasoning.
Their findings showed a significant deterioration in ChatGPT’s performance over time. For instance, in March, ChatGPT-4 demonstrated an impressive 97.6% accuracy in identifying prime numbers. However, by June, this accuracy plummeted to a mere 2.4%, raising concerns about the AI’s reliability.
The researchers also noted ChatGPT’s declining competency in generating code and its newfound tendency to avoid answering certain sensitive questions. AI expert Santiago Valderrama expressed his concerns on Twitter, pointing out that GPT-4 was regressing instead of progressing.
Rumors suggest that the decline in performance might be related to OpenAI’s implementation of smaller and more specialized GPT-4 models, which could be less expensive to run but may compromise overall performance.
Users of ChatGPT have taken to social media and forums to express their disappointment in the declining accuracy and performance when compared to its predecessor, GPT 3.5. Some speculate that the cost of operating these advanced AI systems is hindering the deployment of the best versions to the public.
While OpenAI’s ChatGPT initially made waves, app downloads for ChatGPT and Microsoft Bing, which incorporated GPT-powered technology, have recently experienced a slowdown. A report from the Bank of America, citing Sensor Tower data, revealed that ChatGPT downloads on iPhones in the U.S. dropped by 38% month over month in June. Bing app downloads, featuring a ChatGPT-based chatbot, saw a similar decline of 38% in the same month. Additionally, web analytics firm Similarweb reported a 9.7% decrease in ChatGPT website traffic globally and a 5.7% drop in unique visitors during the same period.
While the generative AI boom remains a significant trend, the declining performance of ChatGPT has raised doubts about the readiness of such technology for widespread adoption. As industry giants like Microsoft and Google continue to integrate AI-powered tools into their core products, the AI community faces increasing pressure to address the challenges posed by these cutting-edge systems.
