ChatGPT & OpenAI

ChatGPT (Chat Generative Pre-trained Transformer) is a chatbot launched in November 2022. 

The answer to how ChatGPT works is in its name, specifically GPT, or Generative Pre-trained Transformer. 

• Generative reflects its ability to produce text that resembles natural human language text
• Pretrained means that the model has been trained on a specific dataset; this includes a large amount of conversational data 
• Transformer refers to the machine-learning architecture 

Altogether this means that ChatGPT works by predicting answers to questions, based and all of the data it has been fed -- similar to studying and cramming before a test a high school math test. However, unlike humans, ChatGPT doesn't change its answers based on feedback, yet.

ChatGPT generates false information

• these tools generate content by predicting the next word based on the training data it's been given. As a result, essays and reports can be filled with fictional citations, quotes and references because the examples it has been given usually have one (ChatGPT and Generative AI, 2023a). 
  • For example: If you ask ChatGPT to write a news story about financial earnings of a company, it'll likely include a non-existent quote from the CEO, cause it knows these stories usually include one.
  • this is commonly known as AI hallucination and can be very convincing and appear natural until you investigate (Metz, 2023). 
• The job of generative AI is not to be 100% right all the time - to sound convincing enough most of the time. 

ChatGPT does not have current information 

• only has access to information from before the last data input (Ortiz, 2023).

Academic Issues

• you cannot use AI tools and present the work as your own
• See this guide's Citing & Academic Integrity page for more info and how to cite ChatGPT in APA. 

ChatGPT is not free from bias

• these tools can be programmed not to answer certain questions that could be harmful, toxic or political 
  • For example: some users have gotten ChatGPT to write positive poetry for Joe Biden, but it refuses to do the same for Donald Trump
• content may also perpetuate harmful biases and stereotypes, depending on the data used to train it (ChatGPT and large language, 2023). 
• While it can be argued that some of these measures have been taken to protect users, they do impart a level of bias that users can't bypass, as that bias reflects that of the developers and engineers creating the algorithm and sourcing it data inputs. 
• ChatGPT uses natural language to generate responses. If there is a bias in the question itself, ChatGPT will recreate this bias in the answer
  • For example: if you ask "Why are socially assistive robots an effective treatment for people with dementia?", ChatGPT assumes that this statement is true and only lists information that supports this thesis (ChatGPT and Generative AI, 2023b). 
• ChatGPT favors Western male perspectives

Privacy Concerns

• We do not know what the company is doing with the data collected from users (Burgess, 2023). 

References:

Burgess, M. (2023, April 4). ChatGPT has a big privacy problem. Wired. https://www.wired.com/story/italy-ban-chatgpt-privacy-gdpr/

ChatGPT and Generative Artificial Intelligence (AI): Incorrect bibliographic references. (2023a, April 27). University of Waterloo Library. https://subjectguides.uwaterloo.ca/chatgpt_generative_ai/incorrectbibreferences

ChatGPT and Generative Artificial Intelligence (AI): Potential for bias based on prompt. (2023b, April 27). University of Waterloo Library. https://subjectguides.uwaterloo.ca/chatgpt_generative_ai/chatgptbias

ChatGPT and large language model bias. (2023, March 5). CBS News. https://www.cbsnews.com/news/chatgpt-large-language-model-bias-60-minutes-2023-03-05/

Metz, C. (2023, March 29). What Makes A.I. Chatbots Go Wrong? The New York Times. https://www.nytimes.com/2023/03/29/technology/ai-chatbots-hallucinations.html

Ortiz, S. (2023, April 18). What is ChatGPT and why does it matter? Here's what you need to know. ZDNet. https://www.zdnet.com/article/what-is-chatgpt-and-why-does-it-matter-heres-everything-you-need-to-know/

In short, ChatGPT is very good at sounding smart. It’s just not very good at being smart.

Check out these examples of ChatGPT getting it wrong:

Copying and pasting an essay topic into ChatGPT seems easy right?

It will produce a chunk of text about that topic, but it's unlikely to be great. This is because ChatGPT and similar AI tools work by giving the tool a prompt, or instructions, to follow. 

A good prompt generates basic text that needs tweaking; a great prompt might need to be run a few times but is specific and useful to the user. For a prompt to be great it needs the following elements: 

• Be specific, clear, and concise. Define what you are looking for. e.g. five ideas for ..., a title of a book about ..., the first sentence of a letter to ... 
• Provide content and avoid ambiguity. Open-ended prompts like 'Write a story about cats' are not narrow enough for ChatGPT to generate a meaningful response. 
• Provide context. If your question is related to a specific topic or field, provide some context to help ChatGPT understand the question. 
• Provide specific constraints, such as a target length or tone, to guide the model's response.  
• Make it conversational and use complete sentences, including proper grammar and punctuation. 
• Additionally, a good prompt should have a clear and achievable goal - answering a question, generating text, translating language, etc. - that the model can work towards.

From: https://libguides.library.qut.edu.au/c.php?g=958007&p=6953208

More Info: