Alarming Trends: What’s Causing ChatGPT’s Capabilities to Decline?

The popular generative AI ChatGPT appears to be getting worse with the passage of time, a new study has revealed. The researchers hailing from Stanford and UC Berkeley discovered ChatGPT’s latest models (Chatgpt 4, 3.5) have been losing their capability of rendering accurate responses to a similar string of questions within a few months time.

However, the researchers weren’t able to offer a precise explanation regarding why the AI chatbot’s performance had declined. The researchers Lingjiao Chen, Matei Zaharia, and James Zou asked ChatGPT-3.5 and ChatGPT-4 models to solve a number of math problems, provide responses to sensitive queries, write new lines of code, and carry out spatial reasoning from different prompts. It was carried out to check the responsive accuracy of chatgpt.

According to research, ChatGPT-4’s ability to identify prime numbers dropped from 97.6% accuracy in March to just 2.4% in June. In contrast, GPT-3.5 improved its prime number identification during the same period. Both models also deteriorated in generating new lines of code between March and June.

In sensitive queries, ChatGPT’s answers shifted from delivering extensive reasoning to being more straightforward in refusing to answer, specifically emphasizing on ethnicity and gender. Moreover, the research sheds light on the significance of continuous monitoring of AI model quality, suggesting users and companies to implement monitoring analysis to keep chatbots effective.

It is pertinent to mention here that OpenAI announced plans for a team to manage potential risks from superintelligent AI, anticipated to be launched around 2030.

Also read:

How to Upload a Document on Chatgpt? Authentic Easy Methods!