Software engineer Eike Hein has put OpenAI's popular yet divisive ChatGPT large language model (LLM) to work building a customized newspaper — displayed on an automatically-updating Espressif ESP32-powered ePaper photo frame.
"Hyepaper an automatic newspaper in a frame, intended as a decorative piece," Hein explains. "It features a 13.3" 1600×1200 E Ink panel in an ESP32-based, battery life-optimized frontend that wakes up once a day from deep sleep to fetch and update with a new newspaper image. All onboard software is written in Rust, including a custom driver for the ITE IT8951 EPD controller chip."
The idea of grabbing and displaying a newspaper's front cover on a paper-like daylight-readable ePaper display isn't new — we've seen multiple examples over the years, including this by Greg Raiz and an even larger one by Max Braun — but the Hyepaper is different: it's created specifically for Hein, using OpenAI's ChatGPT service to summarize the text from multiple configurable news sources.
"The backend running on the [Raspberry] Pi is written in Python," Hein explains. "It includes
lxml-based article scrapers for some websites my wife and I like, and then runs the content through OpenAI's ChatGPT API to trim articles for size, perform light style transfer, and generate suitably short headlines for the layout. It also fetches a weather forecast from the OpenWeatherMap API, which goes into the top-right corner of the header — the layout is of course heavily inspired by the frontpage of the New York Times. With the content in hand it then generates a LuaTeX input file using the Jinja2 template library."
The result looks and reads very convincingly like a newspaper, complete with masthead, imagery, and a columnar format, but is wholly customized. Some smart power handling, including a 5V relay to prevent phantom power drains from the ePaper display driver and a real-time clock which wakes the ESP32 up when it's time to draw a new front page, allows the entire frontend device to run from battery for weeks at a time.