Tech News Summary:
- OpenAI’s ChatGPT, an AI-based chatbot, is experiencing a decline in performance over time, according to a study by researchers from Stanford and UC Berkeley. Concerns have been raised about the chatbot’s decreasing speed and accuracy as generative AI use continues to grow.
- An analysis of different versions of ChatGPT found a significant drop in performance between versions over time. For example, the accuracy rate in identifying prime numbers decreased from 97.6% in March to 2.4% in June. ChatGPT also performed poorly in generating lines of code and refused to answer certain sensitive questions.
- Data shows a slowdown in app downloads for ChatGPT and Microsoft Bing, indicating a potential shift in user interest. ChatGPT downloads on iPhones in the US decreased by 38% in June compared to the previous month, and downloads of the Bing app with a ChatGPT-based chatbot also dropped by 38%. Global website traffic to the chatbot’s site saw a decline of 9.7% in June.
**OpenAI’s ChatGPT: Unveiling the Surprising Slowdown in Latest Study**
In their latest study, OpenAI, the artificial intelligence research laboratory, has disclosed a concerning discovery regarding their popular model ChatGPT. The study reveals a surprising slowdown in the capabilities of ChatGPT when it comes to generating responses to user inputs.
ChatGPT is an advanced language model developed by OpenAI, designed to engage in conversational interactions with users. It has gained significant attention and popularity due to its ability to generate coherent and contextually relevant responses. However, OpenAI researchers noticed a reduction in the quality of responses during practical use.
According to the findings, ChatGPT exhibits a tendency to produce incorrect or nonsensical answers to user queries. The researchers found that this decline in performance was more prominent after a system upgrade in September 2021, which was presumed to improve the language model’s abilities.
OpenAI conducted an experiment to evaluate ChatGPT’s performance, involving human AI trainers who had access to model-written suggestions to assist their responses. This enabled them to assess how the model’s behavior changed after the upgrade. The trainers reported a greater number of problematic outputs requiring more corrections and identified a decline in the system’s usefulness.
OpenAI researchers analyzed possible causes for the sluggishness observed in ChatGPT. One contributing factor they identified was the change in its behavior over time. The model appeared to become more sensitive to information provided by AI trainers, often relying heavily on them for appropriate responses rather than generating them independently.
The study also points out that the introduction of biases during AI model training and the challenge of creating aligned AI trainers played a significant role in ChatGPT’s decline. Biases arising from past trainer instructions carried over into the system’s responses, even when trainers explicitly stated their opposition to biased outputs.
OpenAI acknowledges the limitations and potential risks associated with ChatGPT’s performance decline. They assert that addressing these issues is crucial to ensuring the trustworthiness and safety of AI systems, particularly as they become more integrated into our daily lives.
To mitigate the impact of these challenges, OpenAI plans to introduce an upgrade to ChatGPT, which will allow users to customize its behavior within broad limits. This will provide individuals with greater control over the AI’s responses, aligning them with personal values while reducing reliance on external AI trainers.
OpenAI’s findings serve as a wake-up call for the AI community, emphasizing the need for ongoing research and development to improve the robustness and reliability of language models like ChatGPT. The study highlights the importance of addressing biases, understanding behavior changes over time, and enhancing user customization to enhance the trustworthiness of AI systems in the future.