Create a GPT-3 Lazy Collection

[gregpriday]

We should create a lazy collection that creates new instances of an item using GPT-3. This should be a class that extends the base Larave LazyCollectiont to add methods for adding seed data, a prompt as well as various options.

This is a general-purpose version of #2, in fact, we could even build the model factory on top of this collection.

Options and Settings

These are the options and settings the lazy collection will need.

• Query|Collection|Closure $seed - The seed data that we'll pass to GPT-3. In the case of a closure, we'll call it for each call to GPT-3 so the data is always unique.
• int $chunkSize - we might need to give this a different name. The chunk is how many items we'll get from GPT-3 with each request.
• array $gpt3Options - an array of GPT-3 options.
• Closure $itemToString - a closure that converts an item (usually a seed item) into a string for GPT-3.
• Closure $stringToItem - a closure that converts a string from GPT-3 back into a item (Model, Object, array, etc)
• string $itemSeparator - a string used to separate item instances.
• string $prompt - The prompt string for GPT-3.

Adding Seed Data

The user should be able to set the seed data in a constructor in a separate withSeed.

Using a Closure for Seed data

Here's an example of creating a generator that is seeded with 10 random items from a collection each time its run:

$generator = new GeneratingCollection(fn() => $collection->shuffle()->slice(10));

This, along with a higher temperature setting for GPT-3, should ensure a lot of variety in the results.