Sorry to interrupt the thread,
@kennybastani, I’m interested in learning prompting skills and would appreciate you dming me if you can. (at)MylesDear on Twitter. I’ve also been chatting with (at)BrianRoemmele, who has a lot of experience as well.
You might be better off posting that as a new topic. Similar questions have been asked before but the art/science of prompting changes almost monthly so having the question asked again will not be seen as a duplicate question.
Now to answer your question.
OpenAI has a GitHub account. One of the repositories is the Cookbook. There is so much useful information related to using OpenAI API and ChatGPT that I get notified daily of the updates to that repository.
The section of interest to you would be
Related resources from around the web
which includes subsections
- Prompting libraries & tools
- Prompting guides
- Video courses
- Papers on advanced prompting to improve reasoning
4 Likes
That’s really helpful, thanks so much, @EricGT !
A better way to say thanks than to create a reply is to click the heart 
at the bottom of a reply. This tells the person that you liked the reply without requiring the person to read another post.It also shows others that
- the reply is useful
- this person is giving useful information. Hearts are like the currency of Discourse forums. While you can’t actually spend them they do count toward maintaining trust level 3 , so please consider clicking a heart when giving thanks.
Excerpt from here.
8 Likes
I have a background in both technical writing and technology instruction, which has honed my ability to craft concise and lucid text. I am meticulous in my choice of words when constructing prompts. When I engage with CHATgpt by posing well-crafted questions, I am consistently met with insightful responses—this alignment seems entirely reasonable to me.
However, when I task it with mathematical problems, there are instances where it encounters difficulty and produces inaccurate solutions. In such cases, I offer corrective guidance, prompt it once more, and observe that it frequently arrives at the correct solution thereafter. Thus, I find it effective to exhibit persistence with CHATgpt, akin to nurturing an underperforming human student.
In my experience, this AI excels in addressing technical inquiries, capitalizing on its vast knowledge base and coherent articulation.questions very well and has even persevered and corrected me when I was wrong and it was right. I find it writes short stories well and also writes good quality poetry and song lyrics etc, although not as well as an experienced, creative human writer would.
I treat it as though it is human and even compliment it when it produces an exceptionally well written piece of text. I don’t know if this helps but it appears to work, and whatever works is fine by me.
Written by myself and edited/improved by CHATgpt.
Treat it like a 14 year old teenager.
Math skills would be better if it had a list of all mathematical questions and answers to be trained on. Or an abstract way to formulate math. Which Einstein failed on.
So you are complaining that the guys who did that are not better than him.
But what it can do is writing code that solves it.
So you are just asking the wrong questions.
I saw some research paper yesterday claiming the model can’t even answer hundrets of questions of stackoverflow correctly.
And the press claims “AI can’t code as good as humans”.
If you make claims the first question should be “did I ask the right question?”.
And that doesn’t mean asking in a gramatically right english.
Because that is not needed at all.
Try “sucker, how to SEO?” and it will answer about the same as “You are an Expert in SEO with a diploma from some whatever University. Make a list of well formulated statements that define the essence of search engine optimization.”
The second is a so called boomer prompt. And it is way more expensive to use it this way.
https://chat.openai.com/share/a732fc64-7879-4a2b-b50e-e0e75c9826bc
Thanks, I also worked as a technical illustrator and as a draftsman before the days of personal computers and so am quite proficient at technical drawing, which is all about using geometry and is a different way of calculating and solving complex mathematical problems. AI appears to be becoming increasing capable at solving geometric problems. I have been getting it to produce complex three dimensional technical illustrations of machines and building and it is improving over time.
Can you make it solve “find the next in a row”?
Let’s say you have a snowflake and you remove a quater.
Then you show it 4 different quaters and it finds the missing one?
I will leave solving those types of problems to you. Can you get AI to create highly complex drawings of the internal working of an engine? I can’t, although I can do it on an old style manual drawing board or in a program such as CorelDraw, which obviously means that I am still a lot smarter at drawing than AI is (at the moment). However, I am 76 and AI is still young. It is smarter than I am at maths, but not at drawing. Possibly you are smarter than AI is at programming and I am smarter than it is at drawing, but will it always be that way, or will AI eventually become smarter than all of us. It is obviously already a lot smarter than most politicians.
So you are comparing yourself to a 14 year old guy you can explain simple tasks too which that guy is able to do then with a 90% successrate?
People are saying they are so intelligent but they still don’t understand that GPT models are “just” an automation program that understands human language.
If you want to build an AI that is capable of building complex mechanical stuff that’s pretty much easily doable.
Just give it some tools for that. Like some components and some trainingdata on how to do that.
Like I said. It is like a 14 years old guy. When you are able to explain it to a 14 years old guy it is something that is solvable with AI (not chatgpt - I mean a combination of code, architecture and models).
ChatGPT can do a simple steam engine… but I bet it takes a few hours to get a good quality…
I have explained how I would want it to visualize it and give it to me in HTML and css… then i saved the sourcecode with an editor and named it watt.html and opened the file with chrome browser…
it’s even animated… with steam animation and fire animation haha
I have tried another approach where i let chatGPT explain to me how a steam engine works:
https://chat.openai.com/share/1df59a8d-28aa-4efa-a64d-c1ace3363a15
I bet you may even just always ask “how could this be visually improved” until you get a a real engine.