ChatGPT, the chatbot OpenAI launched in November 2022, managed to go viral all over the world with what it could. ChatGPTis used in many mobile applications today, especially as a player in customer representation.

While we expect ChatGPT to improve with each update, a study published on July 18 has produced some interesting results. Scientists from Stanford and Universities of California agree with every new language model of ChatGPT. begin to give worse answers explained.

Higher failure rate on the latest GPT models:

Research team members Lingjiao Chen, Matei Zaharia, and James Zou put two different versions and two versions of GPT, the language model that ChatGPT created, to the test: GPT-3.5 March/June and GPT-4 March/June.

In the test, various calculation problems and sensitive questions were asked of the language model. Asked to write code and reason about input. Especially the difference between the versions was quite surprising.

The March release of GPT-4 is the release of an issue submitted to it. whether it is a prime number It managed to know with an accuracy rate of 97.6%. In the June version of the same version, this rate is only 2.4% was limited to.

In prime numbers, GPT-3.5’s success was even better than GPT-4 March. The March release of GPT-3.5 releases prime numbers in the same period with higher accuracy he knew. Similar declines experienced as it evolved were reflected in encryption.

Sensitive questions created a strange picture. In March releases, the bot sends a query to the user. why can’t you answer he explained in detail. In June versions, user only he apologizes and he says he can’t answer the question.

The research team also explained the reason behind the decline in ChatGPT’s capabilities. could not determine.