ChatGPT has increasingly conquered users who have been finding different answers to common problems or even for work and study.
However, it is extremely necessary to be careful with the information provided by Artificial Intelligence (AI). The tool, while impressive, is not foolproof.
see more
Big techs plan to launch a 'new Google Maps';…
Claro revolutionizes fixed internet access with the implementation of…
A Purdue University survey tested OpenAI's tool on programming questions, revealing a hit rate of just 48%. The full study is available in this link(in English).
Throughout this text we will analyze what is indicated by the research carried out and understand more details about these alleged major problems involving the ChatGPT.
The survey, conducted by Samia Kabir, David Udo-Imeh, Bonan Kou and Assistant Professor Tianyi Zhang, submitted 517 questions to ChatGPT. Twelve volunteers also participated, evaluating the accuracy of the responses given by the AI.
Challenging results: 48% hit rate
The results were disappointing, with only 48% of the answers correct. Surprisingly, 77% of responses were generic and not easily verifiable. This demonstrates the difficulty in using the tool to solve problems.
(Image: publicity)
The study highlighted that although ChatGPT has advanced, it still struggles with the nuances and subjectivities of the issues.
According to the article, errors are more evident when obvious, but become difficult to identify when they require external and broader knowledge.
Research has found that many erroneous answers are due to the AI's inability to understand the context of the question.
Polite language and well-articulated responses can disguise totally wrong explanations and lead the user into error.
As such, companies like Google have warned about inaccurate information from chatbots. AI, while promising, is not error-free.
It is advisable to use it as an aid, as it still does not completely replace human discernment, causing several problems.
Understanding the limitations of AI, the Purdue University study highlights that ChatGPT, despite its impressive accomplishments, is not infallible in programming matters.
Artificial Intelligence, although evolving, still faces obstacles in understanding complex and subjective contexts.
However, in other types of issues, AI can help when difficulty arises. A useful example of your application is to organize lists, translate texts, among other tasks.
Use the tool wisely and always double-check the answers given to your questions to avoid disclosing possible mistakes made by the tool. AI.