Safely parse and repair JSON returned by GPT-style models

As developers who have worked with large language models like GPT-3, we're all too familiar with the challenges of handling their output. One common issue is parsing and repairing the JSON data that these models return. This can be a daunting task, especially when dealing with broken or malformed JSON.

The problem

When working with GPT-style models, it's essential to handle the JSON output with care. The problem arises when the model's output is not in a standard format, making it difficult for our applications to parse and understand. This can lead to errors, crashes, or even security vulnerabilities if we're not careful.

One common issue is dealing with SRT (SubRip) or VTT (WebVTT) subtitles that may be included in the model's output. These subtitle formats can be complex and require additional processing to parse correctly.…