Do Vision Language Models Understand Human Engagement in Games?

Name: Do Vision Language Models Understand Human Engagement in Games?
Uploaded: 2026-10-03T00:39:49.649884+00:00
Duration: 345 s
Description: Researchers tested if Vision-Language Models (VLMs) can predict human player engagement in games from gameplay videos. Raw predictions were poor, but using few-shot examples improved results significantly. The study found VLMs often rely on superficial visual cues and fail to capture deep reasoning or context, indicating a gap between perception and true understanding.

Researchers tested if Vision-Language Models (VLMs) can predict human player engagement in games from gameplay videos. Raw predictions were poor, but using few-shot examples improved results significantly. The study found VLMs often rely on superficial visual cues and fail to capture deep reasoning or context, indicating a gap between perception and true understanding.

Open the full episode