b2studios
Game development, physics simulations, and reinforcement learning with a focus on visualizing complex algorithms.
Nutrition Label
b2studios builds custom simulations and game engines to visually demonstrate concepts like Reinforcement Learning and physics. The content is highly authentic, prioritizing the engineering journey—including bugs, reward hacking, and failed iterations—over purely theoretical explanations. Viewers get a practical look at how algorithms function within software environments, though the mathematical depth is often simplified for broader accessibility.
Strengths
- +
- +
- +
Notes
- !Expect a mix of polished educational explainers and raw, unscripted stress tests of custom game engines.
- !Check video descriptions for consistent and clear disclosures regarding sponsors and affiliate partnerships.
Rating Breakdown
Breakdown across the key dimensions we rate. Methodology →
Recent Videos
I Forced AI To Learn 4D Movement
How Drones Create Dragons
100 Player Snowball Fight
32 Player Cops And Robbers
Why Aren’t Games Full Of Squishy Things?
250k Special: Q&A
AI Invents New Swimming Stroke
AI Tries Snowboarding (and falls a lot)
AI Plays A Round Of Golf
AI Learns To Play Golf
AI Plays Table Tennis
AI Learns Table Tennis
AI Invents New Bowling Techniques
AI Learns To Swing Like Spiderman
Is It Possible To Win Every Coin Flip?
Why this rating
Evidence receipts showing why each dimension is rated the way it is.
“100 Player Snowball Fight”[0:00] →
The video delivers exactly on the specific promise of the title, documenting the creation and execution of the multiplayer event.
“The AI found a loophole. It realized that if it just jumped off the map immediately, it would get a small penalty for falling, but that was better than the large penalty for existing for a long time.”[3:45] →
Demonstrates a classic Reinforcement Learning failure mode (reward hacking), proving authentic experimentation.
“I gave the AI 150 raycasts... spread out in 4D space. Each raycast returns the distance to the nearest object it hits, or the max distance if it hits nothing.”[2:12] →
Clearly defines the observation space and sensor inputs used for the neural network, demonstrating technical competence.