Improve Handling of Large JSON Data Arrays

[bloodlinealpha]

Overview

The NHL GPT model currently experiences issues when processing large JSON data arrays, leading to the hallucination of data. This issue significantly impacts the accuracy and reliability of the model's output, especially in scenarios involving complex or extensive datasets. Noted by user in issue #1.

Problem Description

When the NHL GPT model processes large JSON data arrays, it tends to generate incorrect or fabricated data, which does not accurately represent the original dataset. This problem appears to be more pronounced with increasing size and complexity of the JSON data.

Problem Analysis

• token window limitations
• python analysis interpreter seems to assume data at times, especially if for larger lists.
• we do not have control of the NHL API, so we cannot modify the response arguments in hopes of reducing the response length.

Expected Behavior

Ideally, the model should accurately interpret and utilize large JSON data arrays without altering, omitting, or fabricating information. The ability to handle complex datasets is essential for maintaining the integrity and usefulness of the model in various NHL data analysis scenarios.

Suggested Enhancement

To address this issue, the following enhancements are proposed:

1. Optimize Data Parsing: Update the GPT configuration logic to put more emphasis on using only necessary data while parsing.
2. Create API to transform data: Create custom API to transform NHL API data and update the GPT Actions to use these more specific endpoints. This would act as a buffer or transformation layer to restructure the data.