FBlick by Nikhil

AI experiments you can watch: small AI models taking on real tasks, starting with classic video games.

On FBlick I test what AI models can actually do, and where they break. Each project follows the full journey: the first attempt that fails, the fixes, the training runs and the result, shown with on-screen decisions so you can see what the AI "sees" and why it acts.

First project: Laya, a 421M-parameter decision model, learns to control the throttle and brake in Sega's Hang-On (1985). Out of the box it never left the start line. After fine-tuning, it finished all 5 stages.

"Blick" is German for "glance". This channel is a glance inside AI decisions.

Built by Nikhil Raj Mule, based in Munich.
LinkedIn: https://www.linkedin.com/in/nikhil-raj-mule-393707a7/