Andrew Ng's New Document Extraction Service: Accuracy Challenges

2025-02-28
Andrew Ng's New Document Extraction Service: Accuracy Challenges

Andrew Ng's newly released document extraction service went viral on X, but Pulse's testing revealed significant issues with complex financial statements, including over 50% hallucinated values, missing negative signs and currency markers. The article argues that such errors can be catastrophic for industries relying on precise data, like finance. Pulse's solution combines traditional computer vision with proprietary table transformer models, achieving higher accuracy and lower latency, addressing the non-deterministic nature, poor spatial awareness, and slow processing speed of LLMs in document extraction.

Read more

LLMs Fail at Complex OCR: Why Large Language Models Struggle with PDFs

2025-02-07
LLMs Fail at Complex OCR: Why Large Language Models Struggle with PDFs

Pulse, a company aiming to extract data from spreadsheets and PDFs, discovered a critical limitation in using Large Language Models (LLMs) for OCR. While LLMs excel at text generation and summarization, they falter significantly when dealing with complex PDFs and tables. The probabilistic nature of LLMs and their abstract image processing lead to hallucinations, data loss, and misinterpretations, posing significant risks, especially with financial and medical data. Furthermore, LLMs are vulnerable to prompt injection attacks, raising security and ethical concerns. Pulse ultimately abandoned LLMs for OCR and is developing a custom solution integrating traditional computer vision algorithms and vision transformers.

Read more
Development