Researchers Unveil ChatGPT Becoming Less Intelligent, How Could This Happen?

ChatGPT Integration Across Various Applications. (photo: Rolf van Root - Unsplash)

In the rapidly advancing realm of Artificial Intelligence (AI), ChatGPT has garnered significant attention as a serviceable chatbot, capable of potentially assuming various human roles. However, recent revelations have taken the tech world by surprise, as contrary to expectations, ChatGPT’s capabilities seem to have dwindled rather than flourished. Developed by OpenAI, this chatbot was designed to continuously learn from vast datasets, leading to the assumption that it would grow smarter over time. But how could it be that ChatGPT’s brilliance has seemingly regressed?

An in-depth exploration into the changes within OpenAI’s ChatGPT, conducted by a team of researchers from Stanford University and UC Berkeley, has brought to light some startling revelations. The underlying language models, GPT-3.5 and GPT-4, were found to exhibit significant variations in performance. Moreover, GPT-4 was discovered to suffer from a continued decline in its capabilities, as reported by Insider on Friday, July 21, 2023.

The ‘intelligence’ of ChatGPT reportedly experienced a dramatic decline within a short span of a few months. In March, ChatGPT demonstrated an impressive 97.6% accuracy rate in identifying prime numbers, but by June, this accuracy plummeted to a mere 2.4%. The chatbot began exhibiting more formatting errors and appeared generally “less willing” to respond to sensitive queries.

Matei Zaharia, a computer science professor at UC Berkeley and one of the researchers involved in the study, opined that OpenAI, as the developer of ChatGPT, faced challenges in managing the quality of the platform.

“The difficulty lies in how well the platform developers detect changes and prevent the loss of ChatGPT’s capabilities,” he remarked.

While the precise reasons behind the decline in ChatGPT’s quality remain uncertain, platforms reliant on data-driven models like AI-powered chatbots necessitate consistent maintenance and learning.

These endeavors undoubtedly come with considerable operational costs. Peter Yang, the product lead at Roblox, speculated that the deterioration in ChatGPT’s quality could be attributed to OpenAI’s desire to cut operational expenses. He claimed that GPT-4’s responses became faster than usual in May, but at the cost of diminished quality.

OpenAI, however, refuted these claims. Last week, Peter Welinder, the VP of Products at OpenAI, asserted that they had indeed created a new version that was smarter than its predecessors. “No, we did not make GPT-4 dumber. On the contrary, we consistently strive to make each new version smarter than the last,” Welinder emphasized in a recent tweet.

As the tech community awaits further developments in this saga, the question of ChatGPT’s true potential and the implications for the future of AI-based chatbots remain at the forefront. The rise and fall of ChatGPT serve as a reminder that the intricacies of AI require ongoing vigilance and thoughtful management to harness the true power of these groundbreaking technologies.