ChatGPT, the immensely popular conversational AI chatbot developed by OpenAI, is experiencing a potential downturn in its popularity and usage. Despite its initial success following its launch late last year, recent data suggest a decline in interest and utilization of ChatGPT. Some longtime users have taken to social media and developer forums to voice their grievances, asserting that the AI’s responses have become of lower quality compared to previous weeks. These users describe ChatGPT as “lazier,” “dumber,” and prone to producing more mistakes and nonsensical answers. OpenAI, on the other hand, denies any intentional downgrading of ChatGPT and claims to continuously enhance its intelligence with each new version.

OpenAI’s GPT-4 AI model gained significant attention earlier this year due to its remarkable capabilities, albeit with the trade-off of slower performance and higher costs compared to other models. GPT-4 possessed the ability to comprehend both images and text, making it the most advanced AI system available at that time. However, speculation suggests that OpenAI might be pursuing a different strategy with GPT-4 by creating multiple smaller models with specific focuses, instead of developing one massive model. This approach, known as a Mixture of Experts (MOE), could help reduce computational costs while providing similar capabilities to a singular large model. By segmenting GPT-4 into specialized smaller models that address narrower tasks, OpenAI can leverage the advantages of a large language model without incurring excessive expenses.

In response to user complaints about GPT-4, Peter Welinder, OpenAI’s product lead, defended the model’s performance on Twitter. Welinder suggested that users may perceive GPT-4 as worse than before because increased usage has unveiled limitations and flaws that were previously unnoticed. He emphasized that each subsequent release of the GPT-4 model is designed to be more intelligent than its predecessor. Despite widespread speculation surrounding GPT-4’s architecture, OpenAI has yet to announce any significant changes.

The perceived decline in ChatGPT’s capabilities aligns with measurable decreases in website traffic and app downloads. According to data from Similarweb, global website visits dropped by nearly 10% in June compared to May. Sensor Tower data also indicates a tapering off of downloads for the iPhone app since its peak in early June. This decline in usage suggests that the initial excitement surrounding AI chatbots may be diminishing as their limitations become more apparent. Other factors contributing to this decline could include the conclusion of the academic year and concerns about regulatory implications. Users fear that chatbots like ChatGPT might propagate false or incorrect information, while companies express concerns about the potential leakage of confidential business data entered into the chatbot.