Tips on Using AI Largely taken from https://canvas.sydney.edu.au/courses/51655 Although the discussion below

Tips on Using AI

Largely taken from https://canvas.sydney.edu.au/courses/51655 Although the discussion below is paraphrased, the vast majority of the information contained herein is from this excellent site, and although missing exact quotations, the ideas and structure came from here.

Prompts

Make your prompts clear. Use as neutral or questioning tone with the AI to avoid an unanticipated bias. AI can be led into giving an answer it “thinks” that the human wants.

Have the AI comment on your prompt. Ask it to help you refine your prompts so that they are clear to both you and the AI. Don’t allow the AI to assume, because it will. You can ask the AI to tell you about any assumptions that it made.

The process will be iterative. Make a prompt and then analyze the output and then try again. Then do it again.

Develop a conversation with the AI. Try being polite. Please and thank you go a long way in human interaction, try the same with AI. Although antidotal, one blog ran a test and seemingly got better responses with a little politeness. Can’t hurt.

More developed prompts will get answers more closely aligned to your needs.

It is not a bad idea to limit the length of the output initially, then as necessary ask the AI to expound on certain topics or limit others as necessary.

Some suggestions on constructing good prompts.

The AI will respond best to a structured prompt.

Role: Define for the AI what its role is. One way to do this is to start out the prompt with “Act as”, “You are”, “imagine”. This changes the way the AI will return information, as well as invites the AI into the discourse.

Task; what do you want the AI to accomplish. Use an outline, be specific and concise. The lengthier you are the more chance to be ambiguous.

Requirements; Give sufficient information for the AI to give you the output desired. Cover all bases and don’t allow the AI much latitude to deviate from the desired material.

Instructions; steps and examples are good. How do you want the AI to go about the task you have given it. Provide the AI constructive criticism on what it gives you.

Seldom will the prompt be done well enough initially, that it won’t be necessary to go back through these steps repetitively to get the desired product.

Good things usually have drawbacks and these need to be taken into consideration.

Just as anything involving the internet, it is a bad idea to provide information that is personal; that not only includes names, addresses, SSI or other specific identification information but includes metadata included with photos, as well as research data, medical data, health information, anything that you don’t want divulged. There are bad people out there that will use that information against you.

Copyrighted material also should be not be uploaded without permission.

It is possible with certain AI to opt out of data collection. ChatGPT is an example. That way your information isn’t stored.

Problems with AI;

Assumptions; given the opportunity, like humans, Ais will make assumptions. Have the AI tell you about them.

Hallucinations; AIs can and do make up facts and citations. Like the internet just because it is there, doesn’t make it right. It is up to you to verify all facts given by the AI. You are responsible, not the AI. Use your wetware.

Bias; The material used in training the AI may have a number of limitations due to lack of diversity, political motivations and be downright prejudicial due the source of the data to train the AI. Garbage in: garbage out.

AI is a series of mathematical predictions or patterns as to what the next phrases might be after a question, and there is really no cognition what-so-ever on the computers part. It would be like Why did the chicken….. and the computer response would be trained …..cross the road. Yet it speaks with authority. You must question that authority. You need to provide the critical thinking necessary to evaluate what the AI product is. Ronal Regan said about the SALT talks “Trust but verify”. In the case of the AI, even trust is problematical.

Although the material used to train AI was licensed, you need to be careful about copyright infringement, plagiarism, citation issues, and other ethical questions.

Ethical test; In order to understand if a submission is ethical to turn in;

Imagine; When you turn in a submission with the help of an AI, are you comfortable and prepared to discuss the submission both the content, as well as the process which the submission was produced, where individual creativity was in evidence, where AI was used, and justify the citations. If the answers to these questions, are yes, then you can feel confident that you have used AI appropriately and your submission is ethical.

Referencing AI assistance

Citation of AI information is not straightforward. Using the exact prompts used, it is very unlikely to reproduce the information that was obtained in a previous session. AI is ever evolving. The best that can be done in in the body of the text, include the “author” tool, version, date, and URL. (See https://canvas.sydney.edu.au/courses/51655/pages/acknowledging-and-referencing-the-use-of-ai?wrap=1 )

Please visit this site. There is still more wonderful information here.

The University of Sydney

“AI in Education”

https://canvas.sydney.edu.au/courses/51655