It had been a frequent question came up to me, about how to keep the research in high intence, yet not too overwhelming so it lower the efficient. This note will explain the steps how I create my own structure of each question, so that I can always get enough information — yet only get deeper to what I want for real!

Note: This note build based on the previous note: How to use a prompt, check that out IF you are confused by how to implement this note pack in practice with Mentiforce!

What is a “good” structure for me?

This is the first question YOU need to think. Sit down, create a markdown file, think what structure is the best… I’ll give you my answer as an example, develop based on this could be a good option:

• Root (Prompt)

Fill content into “{}” will be a realistic full tree that I’d like. You’ll see how there’ll be a stable pattern for each question, that I want ONE AI generated node and ONE me written node. To me, the easier the better.

Hint: You may want AI to create different nodes for you, that you’d like: an example node, AI summary node, AI’s hint, AI’s trick to remember, so on and so forth…

Further improve the quality of output

Given that different item will eventually have different logic cycle to think - we used to “force” ourselves to adapt based on how knowledge came to us. This time though, we get AI to do it for us! That means you need to figure out how your brain explore a topic, and the way to find it would be simple: ask AI to research a topic, then guess the content, then record whatever questions come up to you.

• If you have a topic in mind, great, go for that — otherwise you may also directly ask the AI to find one — “Research any complex [your path related industry] industry problem for me”. In the following example, my question queue looks like this → Spoiler! Try yourself before toggle to see 🙂
• Now you have a basic idea what you want to know - concrete these problems and generalize them - with that being said, you’re an individual, and this is not a course in university, there’s no such a requirement to make you do the whole process by yourself — Use ChatGPT, or even Mentiforce itself, to improve your questions! Also spoiler, try your own result, then compare to what I get: (This also include a brief test in Mentiforce!)

However, this is just a first thought, don’t stuck on to it — when you have new idea, always change it over the time! When you have further confusion, consider if it is something general and if you want add them to the main quesiton pool! Now you’re good to move on to —

Teach LLM to understand

Eventually, we solidified the target, it’s time to teach the LLM how to implement that… And here’s the fun part for me - you can now start to use your creativity to write a paragraph and describe your needs as detail as possible.

LLM is such similar to human (I mean, it do learn from human conversations), that you can use the subjective words like “you need to do…”, “create me…”, ”answer the following questions”;

and as we might not understand the backend tree structure, use blur, general words to make LLM understand your target**,** like “make a section of…”, “creat a block that contains…”;

however, when specify the structure, use the most detail/clear words to describe relations, like “create [a section] directly underneath …”, “put all your answer absolutely parallel to each other”, “add [a prefix] inside of each section, before your title: …”

Undoubtly, trial and error is unavoidable, remember that ctrl+shift+backspace will remove the whole block for you much easier… So I prepared a sample for you so you don’t pay so many time on that — Enjoy!

For each question below, create an "[AI Answer]" section which contains your answer in parellel structure, and also a "[My answer]" section with nothing else, both as direct child of the question. For your [AI Answer] section, only put a short title rather than big paragraph after the "[AI Answer]" prefix, then add some paragraphs underneath the section to explain deeper