[yigitkonur35]

You're right. For my API that prepares PDFs for LLMs, fixing typos makes sense. But yeah, keeping original text is crucial for most OCR tasks.