"Ways ChatGPT Can Fail"
Link to full article:
https://blog.smu.edu/smulibraries/2023/01/20/artificial-intelligence-and-the-research-paper-a-librarians-perspective/
Following are the introductory paragraphs.
AI writing can mimic style, but it cannot mimic substance yet.
The release of a powerful, free and easy-to-use large language model platform, Open AI’s ChatGPT, raises interesting questions about the future of writing in higher education. As the Undergraduate Success Librarian, I have a unique perspective on generative AI, like ChatGPT, that I want to share along with some advice for instructors and students on adapting to AI’s presence in higher education.
What is ChatGPT? How does it work?
ChatGPT is an interface that allows you to interact with artificial intelligence through text inputs and responses. The AI on the other side of the interface is a language model called GPT-3. It produces human-like text by parsing and analyzing the massive corpus of text information (large language) it has been trained on to predict what is likely to come next in a string of words. This makes GPT-3 a type of Generative AI because it uses machine learning to generate new content based on a given set of input data. So, when you give ChatGPT a prompt like “describe losing your sock in the dryer in the style of the declaration of independence” it (in simplified terms) identifies relevant data within its large language dataset, notices patterns within that dataset and then generates a set of text that seems most like the things it identified.*
AI Struggles with Information
By considering GPT-3’s design, we can demystify what it is doing, appreciate what it does well and start to see its limitations, the biggest of which is GPT-3’s inability to really understand what it is saying. The words it produces are statistically plausible, but it is not creating assessments, judgments, behaviors or meaning because it has no internal model or understanding of the topics it writes about. GPT-3 lacks common sense and the ability to reason abstractly. When faced with prompts it has yet to be trained for, it quickly starts fabricating information, making errors or becomes incoherent.
Let’s break down academic writing into 3 categories: structure, style and substance. GPT-3 shows some aptitude with the structural and stylistic elements of writing, but it has some glaring flaws in its use of information – the substance of papers – that make it especially bad at writing college research papers.
End of excerpt from article.
#AI #ArtificialIntelligence #ChatGPT #Chatbot #AcademicIntegrity #TechnologyInTeaching #TippingPoint #DisruptiveInnovation #HigherEducation #AppliedLearning
Link to full article:
https://blog.smu.edu/smulibraries/2023/01/20/artificial-intelligence-and-the-research-paper-a-librarians-perspective/
Following are the introductory paragraphs.
AI writing can mimic style, but it cannot mimic substance yet.
The release of a powerful, free and easy-to-use large language model platform, Open AI’s ChatGPT, raises interesting questions about the future of writing in higher education. As the Undergraduate Success Librarian, I have a unique perspective on generative AI, like ChatGPT, that I want to share along with some advice for instructors and students on adapting to AI’s presence in higher education.
What is ChatGPT? How does it work?
ChatGPT is an interface that allows you to interact with artificial intelligence through text inputs and responses. The AI on the other side of the interface is a language model called GPT-3. It produces human-like text by parsing and analyzing the massive corpus of text information (large language) it has been trained on to predict what is likely to come next in a string of words. This makes GPT-3 a type of Generative AI because it uses machine learning to generate new content based on a given set of input data. So, when you give ChatGPT a prompt like “describe losing your sock in the dryer in the style of the declaration of independence” it (in simplified terms) identifies relevant data within its large language dataset, notices patterns within that dataset and then generates a set of text that seems most like the things it identified.*
AI Struggles with Information
By considering GPT-3’s design, we can demystify what it is doing, appreciate what it does well and start to see its limitations, the biggest of which is GPT-3’s inability to really understand what it is saying. The words it produces are statistically plausible, but it is not creating assessments, judgments, behaviors or meaning because it has no internal model or understanding of the topics it writes about. GPT-3 lacks common sense and the ability to reason abstractly. When faced with prompts it has yet to be trained for, it quickly starts fabricating information, making errors or becomes incoherent.
Let’s break down academic writing into 3 categories: structure, style and substance. GPT-3 shows some aptitude with the structural and stylistic elements of writing, but it has some glaring flaws in its use of information – the substance of papers – that make it especially bad at writing college research papers.
End of excerpt from article.
#AI #ArtificialIntelligence #ChatGPT #Chatbot #AcademicIntegrity #TechnologyInTeaching #TippingPoint #DisruptiveInnovation #HigherEducation #AppliedLearning
RSS Feed