Siraj Raval

“This documentary was sponsored by Copenhagen Atomics. All editorial decisions were mine.”

The creator provides a clear, explicit written disclosure of the sponsorship in the video description.

“I spent 9 hours inside Copenhagen Atomics to see a reactor built for AI's power future”

The creator travels to the facility in Denmark, walks the floor, and physically interacts with the hardware components.

“Every attempt hits a wall... Each one either fails to compile or crashes into some missing piece.”

Shows the reality of failure when attempting the harder problem (Problem 64), adding credibility to the workflow process.

“Pure reinforcement learning. No guides, no human bias... just rewards, punishments, and time.”

The creator claims to use 'pure RL' with 'zero knowledge,' yet the agents navigate complex hierarchical menus (inventory, magic) and read text, a feat requiring advanced computer vision or API access that contradicts the 'from pixels/scratch' claim. The technical implementation for this specific game is glossed over entirely.

“Estimated cost: ~$3/day (~$90/month) for API usage”

The title claims the employees were hired for '$0', but the video's own technical breakdown explicitly lists a monthly operating cost of ~$90 for the necessary APIs.

“I built three AIs, gave them three completely different reward functions...”

While using correct buzzwords (PPO, reward functions), the video fails to explain how the agents handle the enormous action space of an open-world RPG (dialogue, looting, equipping items), which is the primary technical hurdle in such a project.