Yes, the format was specified but it still added some text before the json. By specifying ‘again’ the output format right at the end of the prompt, it seems it helps.
1 Like
Not sure if this has been mentioned before, but you can control the balance between accuracy and creativity and reduce hallucinations in the response. You can specify the desired ‘temperature’ balance in the prompt “Please use a temperature of 0.3 for a balance of creativity and coherence in the response.” The temperature parameter ranges from 0 to 1.
A lower value like 0.1 creates fewer imaginative outputs but is more accurate, while a higher value like 1.0 generates more imaginative but potentially rubbish responses.
2 Likes
Yes, the advice THX pointed out is a great one. Setting the temperature parameter in your prompts is a great way to experiment with integrating different levels of model variability in its responses.
I teach English to Portuguese speakers and I too have been working on helping students (and teaching myself!) become better Prompt Engineers. We came up with this list of Verbs+Nouns+Adjectves which my students (and ChatGPT) and I have found useful:
Prompt Verbs and grammar:**
Revise : Revise and improve the [specific content/document] to enhance clarity/effectiveness.
Arrange : Arrange the following [items/events] in [specific order].
Summarize : Summarize [text/document/article] and provide a concise overview.
Compare : Compare and contrast [two or more entities] in terms of [specific aspect].
Suggest : Suggest possible solutions/ideas/approaches to [specific problem/challenge].
Evaluate : Evaluate the pros and cons of [specific option/decision/strategy].
Predict : Predict the potential outcome/effects of [specific action/event] on [related subject].
Analyze : Analyze the [data/information] to identify patterns/trends/relationships.
Outline : Outline the main points/sections/structure of [document/report/presentation].
Clarify : Clarify the meaning/intent of [specific statement/question].
Hypothesize : Hypothesize about the potential impacts of [specific event/change].
If you would like to copy the list, please feel free, but consider leaving an example of your own? Cheers!
5 Likes
Yep, that is a very good short course, indeed. Just finished!
Interesting. Can AI be programmed to assume, scan and learn context without explicit prompt writing? Text is a consecutive encoding of information that humans developed. But humans don’t communicate whole context every time. They program each other using verbal and text communication and also verbal body language, but they use tiny bits to communicate when the context is already synced. The problem is to sync and recall the context. Like we do with friends. But if my tool to sync with friends is to “drink” in a bar, how to see shared context and synchronize it with friends like ChatGPT?
I think the first 3 in particular are quite helpful and I would take that advice 
Did you try it?
Here are two ideas from the 31 May 2023 New York Times online version.
1. Act as if Beginning your prompt with these magic words will instruct the bot to emulate an expert. For example, typing “Act as if you are a tutor for the SATs” or “Act as if you are a personal trainer” will guide the bots to model themselves around people in those professions.
2. Apply a framework She [a person quoted previously in the article] recently read “Clockwork,” a book about setting up a business. When she asked ChatGPT-the-business-coach to offer advice using the framework of “Clockwork,” she was delighted to see that it could incorporate principles from the book into an action plan for expanding her company.
Do these work for you?
1 Like
Thanks for mentioning this. The temperature parameter for the API was covered with some good explanation and examples in the short course, but I was just thinking about how to apply that to the Web UI. Of course simply telling ChatGPT is the answer!
1 Like
One bit that I like to add to my prompts is who the audience is, or what my current level of understanding is on a topic. For example, I might add something like, “Please express the result in terms that an entry level non-technical employee will understand.”
2 Likes
personality and use of framework is such a ninja technique, like you could use business framework like the lean canvas and it would give you everything according to it!
1 Like