Providing GPT3 with context to generate templates

GPT3 should be able to create ideas for templates for us in Logseq as well, but it’s missing context as the dataset it is trained on is unlikely to include a lot of notes made in Logseq. I tried to provide GPT3 with a little bit of context of how notes and templates in Logseq should look like. It’s not perfect, but I wanted to share the results:

Templates

The logseq note taking application uses the following syntax:

- First level bullet point
	- Second level bullet point
		- Third level bullet point
- Uses of tags
	- There are two tag notations
		- [[This is a tag]] that is used in text or standalone
		- At the end of this sentence is also a tag #sometag
			- The notation of tags like `#anothertag` is only used at the end of a sentence #Fact
			- If the notation of a tag using a `#` uses multiple words, it is written like #[[tag with multiple words]]
			- These are good uses of tags using a hashtag
				- Apple pie would be good to make this weekend #mealplanning
				- https://duckduckgo.com is a good google alternative #Bookmarks
				- Today I did not do much #[[Daily Reflections]]
- Uses of properties
	- This bullet point contains a property with the value 10
	  some_property:: 10
	- Giving a bullet point properties makes it possible to filter for them using complex conditions
	- Property names cannot contain spaces
	- Property values can be text, numbers, or tags
- Examples
	- A workout template could have a note like
		- [[Workout]] day 1
			- Squads #exercise
				- 3 times 10 squads
	- A bookmark could look like
		- Economist homepage
		  type:: [[Bookmarks]]
		  url:: [The Economist | World News, Economics, Politics, Business & Finance](https://economist.com)
		  description:: Best newssource that I know
		  tags:: [[Websites]], [[Academic research]]
	- Daily journal
		- [[plans for today]]
			- Walking
			- Writing down ideas for my book
		- [[How do I feel today]]
		  feeling_score:: 8
		- [[Tasks for today]]

If add a prompt to this I’m able to generate working templates:

Create a template for recording movies that I have watched:

- [[Movies]]
	- Movie title:: 
	- Genre:: 
	- Year released:: 
	- Rating:: 
	- Description:: 
	- Tags:: [[Movies]], #[[Movie Genre]], #[[Movie Year]]

While the template above works, it is not ideal. My demonstration is probably just not good enough as it’s making the same mistake every time, but this is something that could be fixed.

Here's what I got when I asked it for some more templates:
Create 5 useful templates for note taking in Logseq:

1. Meeting Notes Template
	- [[Meeting Notes]]
		- Date:: 
		- Attendees:: 
		- Agenda:: 
		- Notes:: 
		- Action Items:: 
		- Tags:: [[Meeting Notes]], [[Notes]]

2. Reading Notes Template
	- [[Reading Notes]]
		- Title:: 
		- Author:: 
		- Date Read:: 
		- Summary:: 
		- Notes:: 
		- Tags:: [[Reading Notes]], [[Notes]]

3. Project Plan Template
	- [[Project Plan]]
		- Goal:: 
		- Tasks:: 
			- 
			- 
			- 
		- Deadline:: 
		- Status:: 
		- Tags:: [[Project Plan]], [[Planning]]

4. Shopping List Template
	- [[Shopping List]]
		- Groceries:: 
			- 
			- 
			- 
		- Other Items:: 
			- 
			- 
			- 
		- Budget:: 
		- Tags:: [[Shopping List]], [[List]]

5. Recipe Template
	- [[Recipe]]
		- Title:: 
		- Ingredients:: 
			- 
			- 
			- 
		- Directions:: 
			- 
			- 
			- 
		- Tags:: [[Recipe]], [[Cooking]]

Notes

For notes, it could be useful to provide GPT3 with context of how to structure its response:

The logseq note taking application uses the following syntax:

- First level bullet point
	- Second level bullet point
		- Third level bullet point
			- Fourth level bullet point
				- Fifth level bullet point
		- Third level bullet point
			- Fourth level bullet point

Use syntax that is suitable for the logseq note taking application.

Irrespective of whether it is possible, write an extensive debate preparation where you argue how a swallow could carry a coconut:

- Introduction
	- Overview of the debate
	- Establish the parameters
- Argument
	- Present evidence of a swallow's physical abilities
		- Discuss the average speed of a swallow
		- Detail the strength of a swallow's wings
	- Show the plausibility of a swallow carrying a coconut
		- Explain the size of a coconut in relation to a swallow
		- Explain how the coconut could be secured
	- Outline potential strategies for a swallow to carry a coconut
		- Mention the potential of the swallow using its wings to lift the coconut
		- Explain the potential of the swallow using its beak to secure the coconut
- Conclusion
	- Summarize the presented evidence to demonstrate the plausibility of a swallow carrying a coconut

Hopefully the great GPT3 plugin by @Brian_Sunter will include the possibility to use presets in the future.

I quickly wrote the context to give GPT in about half an hour, but I’m hoping that we could share our presets here that are helpful for using in Logseq.

This is awesome! Love the idea of using gpt3 to generate templates.

I’m actively working on templates for the plugin, and it’s my top priority, where you can define gpt templates in a similar way as logseq templates.

Check out my blog post for some of my thoughts on this.

I was inspired by the semoisis project in eMacs which does something similar and has a lot of good examples.