Library: Tunxis Generative Artificial Intelligence (AI): Generative AI Reliability and Authority

When A.I. Chatbots Hallucinate

New York Times Reporters Karen Weise and Cade Metz asked ChatGPT,

When did the New York Times first report on "artificial intelligence"?

According to ChatGPT, it was July 10, 1956, in an article titled “Machines Will Be Capable of Learning, Solving Problems, Scientists Predict” about a seminal conference at Dartmouth College.

The 1956 conference was real. The article was NOT!

Weise and Metz further warn,

ChatGPT simply made it up. ChatGPT doesn’t just get things wrong at times, it can fabricate information. Names and dates. Medical explanations. The plots of books. Internet addresses. Even historical events that never happened.

Uses and Limitations of AI Tools

Uses	Limitations
`Practical Applications` How to use AI to do practical stuff: A new guide (Ethan Mollick \| One Useful Thing Blog) Write stuff Make images Come up with ideas Make videos Coding Learn stuff	`Hallucinations & Fabricated Sources` Fabrication and errors in the bibliographic citations generated by ChatGPT (Scientific Reports). The difficulty is inherent in large language models, which "use deep neural networks to predict the next word in a sequence of text and provide responses based on statistical patterns learned during training…. As such, ChatGPT cannot distinguish between accurate and false information." Artificial Hallucinations in ChatGPT: Implications in Scientific Writing (Alkaissi & McFarlane \| PubMed Central) While ChatGPT can write credible scientific essays, the data it generates is a mix of true and completely fabricated ones. This raises concerns about the integrity and accuracy of using large language models in academic writing, such as ChatGPT. When ChatGPT writes scientific abstracts, can it fool study reviewers? (Marla Paul \| Northwestern Now) Scientists can be fooled, the study reports. Blinded human reviewers — when given a mix real and falsely generated abstracts — could only spot ChatGPT generated abstracts 68% of the time. The reviewers also incorrectly identified 14% of real abstracts as being AI generated.
`Some Accessibility Gains` Transcriptions, Screen Readers, and ChatGPT (Ben Brumfield \| From the Page) AI might be a real help for making verbatim texts more accessible to the blind and visually impaired. Using AI for increasing access to our texts (rather than cheating on essays or spamming people) gets me excited. An Autoethnographic Case Study of Generative Artificial Intelligence's Utility for Accessibility (Kate S. Glazko, Momona Yamagami, et al.) Our work demonstrates the potential for GAI today to be used by people with disabilities to provide on-demand support for their accessibility needs, in low-stakes, easilyverifiable contexts.	`Privacy and Profit` Privacy Policy (OpenAI) ChatGPT's Privacy Policy states that data collected from users may be shared with third party vendors, law enforcement, affiliates and other users. Some privacy policies may allow the AI Tools and Large Language Models to profit off of your information and interactions. What is a “corpus”? And why is everyone in AI suddenly talking about it? Here’s what you need to know (Michael Grothaus \| Fast Company) “[The ChatGPT corpus] consists of a wide range of text from the internet, including websites, books, articles, and other publicly available sources,” Prompts and data that users submit to the AI Tools may become part of the Large Language Model's training and corpus of learning.

Uses	Limitations
`Practical Applications` How to use AI to do practical stuff: A new guide (Ethan Mollick \| One Useful Thing Blog) Write stuff Make images Come up with ideas Make videos Coding Learn stuff	`Hallucinations & Fabricated Sources` Fabrication and errors in the bibliographic citations generated by ChatGPT (Scientific Reports). The difficulty is inherent in large language models, which "use deep neural networks to predict the next word in a sequence of text and provide responses based on statistical patterns learned during training…. As such, ChatGPT cannot distinguish between accurate and false information." Artificial Hallucinations in ChatGPT: Implications in Scientific Writing (Alkaissi & McFarlane \| PubMed Central) While ChatGPT can write credible scientific essays, the data it generates is a mix of true and completely fabricated ones. This raises concerns about the integrity and accuracy of using large language models in academic writing, such as ChatGPT. When ChatGPT writes scientific abstracts, can it fool study reviewers? (Marla Paul \| Northwestern Now) Scientists can be fooled, the study reports. Blinded human reviewers — when given a mix real and falsely generated abstracts — could only spot ChatGPT generated abstracts 68% of the time. The reviewers also incorrectly identified 14% of real abstracts as being AI generated.
`Some Accessibility Gains` Transcriptions, Screen Readers, and ChatGPT (Ben Brumfield \| From the Page) AI might be a real help for making verbatim texts more accessible to the blind and visually impaired. Using AI for increasing access to our texts (rather than cheating on essays or spamming people) gets me excited. An Autoethnographic Case Study of Generative Artificial Intelligence's Utility for Accessibility (Kate S. Glazko, Momona Yamagami, et al.) Our work demonstrates the potential for GAI today to be used by people with disabilities to provide on-demand support for their accessibility needs, in low-stakes, easilyverifiable contexts.	`Privacy and Profit` Privacy Policy (OpenAI) ChatGPT's Privacy Policy states that data collected from users may be shared with third party vendors, law enforcement, affiliates and other users. Some privacy policies may allow the AI Tools and Large Language Models to profit off of your information and interactions. What is a “corpus”? And why is everyone in AI suddenly talking about it? Here’s what you need to know (Michael Grothaus \| Fast Company) “[The ChatGPT corpus] consists of a wide range of text from the internet, including websites, books, articles, and other publicly available sources,” Prompts and data that users submit to the AI Tools may become part of the Large Language Model's training and corpus of learning.

Tunxis Campus Library

Tunxis Generative Artificial Intelligence (AI)

When A.I. Chatbots Hallucinate

Uses and Limitations of AI Tools