Why basic Maths skills of ChatGPT are declining?

In November of 2022, ChatGPT was launched, capturing the attention of people who found diverse applications for it, such as writing, emailing, creating personal chatbots, coding, scripting songs, and more.

With the release of ChatGPT 4 in March, AI tools once again became viral on the internet due to their enhanced capabilities.

However, some researchers have raised concerns about the foundational Math skills of ChatGPT.

Research conducted at Stanford University and the University of California revealed that while the development of complex AI models has progressed, ChatGPT's performance in basic maths has suffered, reported Live Mint.

James Zou, a Stanford professor and one of the authors of the research, noted that attempting to improve the AI in one aspect can lead to setbacks in other aspects, making consistent improvement a challenge.
 
According to the study, ChatGPT exhibits a mix of performance. While it demonstrates proficiency in certain cases, it often struggles with basic maths, leading to incorrect answers.

Users who engage with both ChatGPT 3.5 and 4 have reported an increase in inaccuracies and instances where the chatbot provides irrelevant responses.

In response to inquiries about the research, OpenAI provided a written statement acknowledging their commitment to enhancing newer model versions across a range of tasks.

They acknowledged that their evaluation methodology is a work in progress, but they are actively working on improving it.

It is important to note that ChatGPT's performance has not universally declined. There are areas where GPT-3.5 has shown improvement, despite its lower overall accuracy compared to GPT-4 in certain cases, as per the report.