Score: 0
Lines: 0
Generation: 0
Population Size: N/A
Max Generations: (0 = infinite)
Best Score Ever: 0
Best Weights Found:
N/A
Current AI Heuristics:
N/A
I played around with the new Gemini 2.5 Pro Experimental (03-25 model date), and it's an amazing model for writing code – so far, it has solved tasks of almost any complexity I've given it; no, it won't build a time machine or create Vice City from a single prompt – but things like shaders, small 3D projects, HTML/React applications, etc. – easy peasy.
In the demo, Google's new model made a genetic Tetris simulator that teaches itself to play by optimizing its strategy – it turned out quite complex: a single HTML file contains the Tetris game itself, an AI player deciding where to place pieces based on a set of rules, and a genetic algorithm, all written from a single prompt.
The genetic algorithm's task is to find the ideal "weights" for these rules so the AI plays as efficiently as possible. How does it learn?
And the coolest part – you can watch in real-time on the screen how the best AI of the current generation tries to break records, and see in the info panel how its weight coefficients change – the very "strategy" it has learned.
And the best thing is that it's all free: https://aistudio.google.com/app/prompts/new_chat