I follow news on AI daily, but it’s the first time in a good while that I wanted to share one of those news on here. Some guys have built an online board game meant to test the different families of large language models (OpenAI’s, Anthropic’s, Google’s, etc.) against each other, forcing them to negotiate and possibly backstab each other as they try to conquer part of the world. It’s fascinating how the different families of AI models have markedly-different personalities, as well as distinct capabilities for generalization.

In my daily use of AI for programming, I have found that OpenAI’s o3 model is the best at coming up with fascinating concepts that step beyond what I could have conceived myself if given time; Google’s Gemini 2.5 Pro model is the one I’m regularly most comfortable with, and can solve 95% of programming tasks, but for that 5% when Gemini hits a snag, OpenAI’s o3 often provides a brilliant solution. Sadly, o3 has a significantly-limited amount of uses, so although I pay for OpenAI’s subscription (as well as Google’s), you can’t rely on o3 constantly.

Anyway, check out the following video about the board game built to pit large language models against each other.

Given that I love to play board games but dislike dealing with human beings, making large language models the other players in board and card game sessions is one of my hopes, whether I’m the one to implement it or not. Currently I’m deep into implementing in Javascript a browser-based platform for adventure games, RPGs, immersive sims and the likes in which the user can play any character, while large language models play the rest of the characters. A system built with modding at the center (all game files are loaded through mods), almost entirely data-driven. Working on it is a joy.