ChatGPT may be the best AI so far to be used by people all around the globe and with its latest merge with Microsoft in the Bing search engine, it shows so much prospects at reduced cost. Kevin Roose, author of “Futureproof: 9 Rules for Humans in the Age of Automation,” has praised the ChatGPT company, OpenAI’s widely acclaimed large language model as “quite simply the best artificial intelligence chatbot ever released to the general public,” and Nvidia CEO Jensen Huang has called it “one of the greatest things that has ever been done for computing.”
Some people think ChatGPT has successfully passed the Turing test, a long-standing test of a machine’s capacity to equal human intelligence. This is because ChatGPT has gotten so excellent at responding naturally to user enquiries. In a variety of subjects, ChatGPT has achieved the greatest percentiles of accomplishment, including math (89th), law (90th), and GRE verbal (99th).
And early in July 2023, researchers from NYU’s medical school stated that the guidance supplied by ChatGPT for health-related issues was nearly identical to that given by actual medical staff. But scientists from Stanford University and the University of California, Berkeley aren’t yet ready to give ChatGPT any responsibility for making important decisions. Lingjiao Chen, Matei Zaharia, and James Zhu noted ChatGPT performance has not been constant, echoing an increasing number of worries recently voiced by users. It is getting worse in certain cases.
Researchers stated in a report that was published on July 18 on the arXiv preprint site that “performance and behavior of both GPT-3.5 and GPT-4 vary significantly” and that responses on some tests “have gotten substantially worse over time.” They saw significant performance differences from March to June, a four-month span. The researchers concentrated on a few topics, such as computer code production and solving mathematical puzzles. In March 2023, GPT-4 solved issues involving prime numbers with a 97.6% accuracy rate. The Stanford researchers found that when the June 2023 model was employed, the rate fell to just 2.4%. For its capacity to help programmers with programming and debugging challenges, ChatGPT has received high accolades. In response to demands from coders, GPT-4 completed correct, ready-to-run scripts a little over 50% of the time. But by June, the rate dropped to 10%. Chat-GPT-3.5 also showed a notable decline in accuracy, from 22% in March to 2% in June.
Conclusion
Even the development community in Stack overflow has banned ChatGPT answers. The same reason, they are scared of the numerous errors the AI can commit. And although, it is a tool to make workflow easier, it is still very much prone to errors. Let us know if you think ChatGPT is airtight or still need more work done on it in the comment below.
