ChatGPT, the viral conversational AI chatbot created by OpenAI, seems to be losing some of its initial luster and appeal.
Having achieved immense popularity after its release late last year, recent data indicates that use and interest in ChatGPT may be declining.
Some long time users have it complained on social media and developer forums that the AI seems to be producing lower quality responses than it was just a few weeks ago.
They describe the bot as “lazier”, “dumbder” and prone to more errors or nonsensical responses. However, OpenAI denies intentionally downgrading ChatGPT, tweeting that “we make every new version smarter.”
The company speculates that users encountering more defects reflects increased usage that uncovers limitations.
From accurate and expensive to fast and wrong
OpenAI’s GPT-4 AI model gained fame earlier this year for its remarkable abilities, even though it was slower and more expensive than other models.
It could understand images and text, making it the most capable AI system at the moment.
It is speculated that instead of developing one massive GPT-4 model, OpenAI is creating several smaller GPT-4 models focused on a specific topic.
This approach, called Mixture of Experts (MOE), would reduce the computational costs of the system while providing capabilities similar to a single giant model.
By splitting GPT-4 into smaller specialized models focused on narrow tasks, OpenAI could benefit from a large language model without such expense.
Answer from OpenAI
After people complained about GPT-4, Peter Welinder, product manager at OpenAI, defended the model on Twitter.
He said users might think GPT-4 is worse than before because they use it more and see problems they hadn’t noticed before.
He insists that each new version of the GPT-4 model is smarter than the previous version.
Despite widespread speculation about the architecture of GPT-4, OpenAI has not announced any significant changes.
Declining user base
This perceived decline in ChatGPT’s capabilities aligns with a measurable decline in website traffic and downloads.
Data from the Internet Analytics Company Similar website shows that visits to the global website were down almost 10% in June compared to May.
Downloads of the iPhone app have also fallen since peaking in early June, data shows Sensor tower.
The decline in usage suggests that the initial enthusiasm for AI chatbots may wane as their limitations become more apparent.
Other factors such as the end of the academic year and regulatory concerns could play a role in reduced use.
People are concerned that chatbots like ChatGPT may create false or incorrect information, while businesses are concerned that confidential business information may be leaked if entered into the chatbot.
To sum up
ChatGPT’s declining user base and worsening perceived performance suggest the viral chatbot may struggle to sustain its initial momentum.
This trend highlights the most considerable challenges facing chatbot creators: as advanced as systems like ChatGPT are, they lack human context, common sense, and reasoning.
Still, ChatGPT represents a significant leap forward in natural language processing, even if its abilities fall short of some people’s imagined potential.
Going forward, developers will need to balance management expectations as they strive for the next evolution of AI chatbots.
Featured image: Vitor Miranda/Shutterstock
[ad_2]
Source link
