With advances in deep learning and image recognition, computers are getting better at properly identifying text and objects in images. There have been several interesting papers and projects applying these deep learning image recognition methods to CAPTCHAs. One Python-based project (https://github.com/arunpatala/captcha) uses PyTorch to train a solver model on a large dataset of CAPTCHAs. In June 2012, Claudia Cruz, Fernando Uceda, and Leobardo Reyes (a group of students from Mexico) published a paper with an 82% solving accuracy on reCAPTCHA images (http://dl.acm.org/citation.cfm?id=2367894). There have been several other research and hacking attempts, especially those targeting the often-included audio components of the CAPTCHA images (which are included for accessibility purposes).

It's unlikely that you'll need more than your OCR or API-based CAPTCHA-service to solve CAPTCHAs for the web scraping you encounter, but if you are curious to try and train your own model for fun, you will first need to find or create a large dataset of properly decoded CAPTCHAs. Deep learning and computer vision are rapidly-advancing fields, and it's likely that even more research and projects have been published since this book has been written!