Barebones API wrapper for reading a prompt from JSON and writing back the assitant response into the same JSON file, to build up a conversation with context.
Put a prompt in the 'content' property of a message with role: user, then run npm start (or node built/chat/app.js). Reload JSON, repeat...
The file chats/messages.json can be used as a template.
The script will make a request to OpenAI API for each JSON file in the chats/ dir where the latest message in the array is a user prompt.
Used the chat/app.ts template to write a really bad demographics model. The file messages.json is the full conversation (user prompts and assistant responses).
Serve barchart.html and demographics.json on localhost, eg by running python -m http.server in the project directory.
Then go to http://localhost:8000/barchart.html
- provide a way to "scroll through" annual data year by year.
- run multiple sims and compare outcomes
- apply scenarios, eg changing fertility rate, death rates etc
- model number of people requiring elder, # providing, length of time to train new practitioners, training capacity, cost to grow training capacity etc.
Playing with dall-e for text to image instead of LLM chat models.
Example usage.
Start a new dir to keep things "tidy":
mkdir 20230522-02
Create prompt parameters in 20230522-02/createImage.params.json, eg:
{
"prompt": "A flying cat with wings, in the style of a childrens book illustrator. Bright colours, simple shapes, acrylic",
"n": 10,
"size": "256x256",
"response_format": "b64_json"
}Run the script:
node built/pictures/pictures.js 20230522-02/createImage.params.json
PNGs created at project root, move them to "tidy" dir. :-)
- write some HTML for prompt + image results, and serve that up -- each time the script gets run, collection of prompts/results grows.