ChatGPT 5.5 Pro simulates chess game, generates multi-page analysis PDF
ChatGPT 5.5 Pro (Extended) simulated a full chess game and output a multi-page PDF report with move-by-move analysis. The model reimplemented quiescence search, evaluated its own play, and embedded a QR code linking to the Lichess PGN viewer. Riley Goodside, Staff Prompt Engineer at Google DeepMind, ran the demonstration to test chess reasoning and report-generation capabilities. Researcher Will Depue replied that a decent chess engine could fit inside a QR code.
First post
Why it matters
Will Depue (OpenAI member of technical staff) built an instructional chess engine in Golang rated above 2100 Elo.
Riley Goodside (Google DeepMind staff prompt engineer) observed that chess receives little attention during large language model post-training.
Quiescence search lets chess engines explore captures and checks past their normal search depth to avoid the horizon effect.
![Screenshot of ChatGPT dialog
User:
Wood-framed print in a modern art museum of a QR code that uses the lichess PGN viewer to display a game where both sides play entirely at random for 67 half-moves and then draw by mutual agreement.
ChatGPT:
> Thought for 2m 7s
[Generated image matching above description.]](https://pbs.twimg.com/media/HGttrM_WkAAQ95a.jpg)





