Show HN: A competition for small neural networks that play strategy games
TinyBrains revives the beloved 2011 Google AI Challenge 'Ants' game, but with a cunning modern twist that pits neural networks against each other. This time, the ultimate flex isn't just about building the most powerful AI player, but about achieving peak performance with the absolute smallest 'brain' in terms of bytes. It's a delightful blend of retro gaming, cutting-edge AI, and fierce optimization, perfectly crafted to pique the interest of HN's technically-minded competitive spirits.
The Lowdown
TinyBrains introduces an engaging new artificial intelligence competition that re-imagines a classic challenge from over a decade ago. It resurrects the strategic depth of the 2011 Google AI Challenge's 'Ants' game, bringing it into the era of trained neural networks.
- The original 'Ants' game, developed by the University of Waterloo Computer Science Club, involved managing colonies in a fog-of-war environment to gather food and destroy rival ant hills, with all colonies moving simultaneously.
- TinyBrains faithfully reimplements these core game rules, but shifts the participation from writing direct bot logic to training and deploying neural networks.
- The unique twist of this new competition lies not in building the strongest AI player, but in creating the most effective player within the smallest possible neural network footprint.
- Submissions are categorized into various 'weight classes' based on the byte size of their neural network, emphasizing efficiency and compact design over raw computational power.
This innovative competition challenges participants to push the boundaries of efficient AI design, demonstrating that sometimes, the smartest solution comes in the smallest package.