AI Use Case Update: Newsletters and Advanced Prompting

On June 3, I suggested using a LLM as a way to gather activities for a newsletter for parents. Mike Caulfield in his The End(s) of the Argument substack takes this one step farther with advanced prompting to promote better results:

Anyway, to have it pull all the data I need while I get to be as lazy as possible I used a little pseudo-code technique I call “LLM paging” where you embed a command in the prompt that can later be used to iterate over a range of data.

AI Generated Juicing

I think in the near future we will get LLMs that are more comfortable doing these lengthy loops without stopping for human feedback, but for the time being this pseudo-code can tide you over.

Here is the code for grabbing events:

# Base Instructions 

Search for cultural events in Europe happening week {#1} of July, and put them in a table with the date a description and a link. Use provided tips below. 

There should be at least 34. Categories might include music festivals, cultural celebrations, specialty/quirky events, folkloric stuff, pride events, arts and performance, food and lifestyle, maritime and aviation. Check table for any duplicate entries and remove. 

## Tips on date searching 
When searching for events in certain date ranges, it is helpful to use "OR" logic with a series of specific dates, e.g. instead of searching for "july 3-july 9" search "july 3" OR "july 4" OR "jul 5" etc. within the date range. 

Be aware of linguistic differences when searching dates, both language and formats. 

## Tips on searching 
Remember the specific links you find talking about each festival, you will use these links later in the table.

# Templates (run based on certain phrases/"hotkeys") 

## Template hotkey="double check" 
Look at each entry in the table and double-check dates, location, and links, and report any errors 

## Template hotkey="next" 
Replace {#1} in instructions with the next number (e.g. week 1 becomes week 2, week 2 becomes week 3)

By combining the vibe coding capabilities of the llm with the search functions released earlier this spring - you "teach" the llm how to a deeper and more thorough search.

A great innovation!