Andrew Ng's newly released document extraction service shows significant limitations when processing complex financial statements, with high error rates and slow processing times. Tests revealed over 50% hallucinated values and frequent missing data in financial tables, highlighting the challenges of using LLMs for document extraction.
A new benchmark evaluates Vision-Language Models against traditional OCR systems for text recognition in video environments, using a dataset of 1,477 annotated frames from diverse sources. Advanced models like Claude-3, Gemini-1.5, and GPT-4o demonstrate superior performance in many scenarios, though challenges with hallucinations and occluded text persist.
OCR4all provides a completely free, open-source optical character recognition solution without any paywalled features or private code restrictions.
A technical exploration demonstrates how Unicode variation selectors can be used to encode arbitrary data within any Unicode character, making it possible to hide invisible messages in text or emojis while surviving copy/paste operations.
An extension to regular expressions called 'transductive regular expressions' (trre) introduces a new ':' symbol for pattern matching and text modification, offering a more natural approach to text editing than traditional regex. The implementation includes a command-line tool that supports various operations like replacement, deletion, and insertion, utilizing Finite State Transducers for processing.