According to researchers at Stanford University and Berkeley, ChatGPT’s intelligence is plummeting. The chatbot would be more “stupid” with a decline in the quality of its answers in many areas over a very short period of time.
ChatGPT improves with updates. Soon, the chatbot will be able to remember who you are thanks to a contextualization feature. But according to researchers at Stanford and Berkeley, the conversational agent does not improve. On the contrary, his intelligence is declining, the quality of his answers is in free fall.
Falling quality of responses in a short time
According to a study published on Tuesday 18 July, ChatGPT continues to lose intelligence. According to the researchers, the conversational agent’s responses declined in quality between March and June 2023.
The researchers came to this conclusion by comparing the relevance of GPT-3.5 and GPT-4 responses. Questions were asked in several areas such as mathematics, coding and even logical reasoning. The researchers also addressed sensitive topics. According to them, “the behaviour of ChatGPT-3.5 and GTA-4 changes dramatically over a very short period of time.”
Proof of the huge decrease in the precision of the answers, ChatGPT-4 was 97.6% accurate in the indentification of prime numbers in March 2023 against 2.4% in June 2023. On the other hand, ChatGPT-3.5 showed better performance between the first and second periods.
For coding, errors increased for ChatGPT-3.5 and GPT-4 between these two periods. Same for responses to sensitive questions.
In both cases, errors and inaccuracies increase, proof that artificial intelligence loses in quality over time? As a reminder, its popularity has been at half-mast for several weeks.
What are the recommendations for this drop in quality?
Faced with this decline in the quality of ChatGPT’s responses, researchers recommend that companies that have integrated artificial intelligence into their models be cautious. Errors are creeping in and such a decline in the relevance of information in such a short period of time warrants increased monitoring.
OpenIA denies that ChatGPT is more stupid in a tweet published before the release of this study. According to one official, the answers are less specific because… you use artificial intelligence much more than you did in the past and you notice mistakes that were once unnoticed.
Post a Comment