Presentation
Can AI Judge Its Own Work? Self‑Assessment and Trust in Creative AI Pipelines
DescriptionGenerative and agentic AI increasingly grades its own output. VLM and LLM judges decide when a result is good enough, confidence scores gate quality, and autonomous tools self‑terminate. But when can we trust an AI's assessment of its own work? This facilitated discussion gathers researchers and practitioners across graphics, vision, HCI, and production to compare notes on AI self‑assessment and calibration. We will discuss where self‑scores break down, how overconfidence hides in iterative loops, and which external‑validation strategies work (second‑model judges, geometric checks, human review). Bring your own failure cases. Researchers, engineers, and students welcome.
Organizer
Event Type
Birds of a Feather
TimeMonday, 20 July 20269:30am - 10:30am PDT
LocationRoom 510
Arts & Design
Gaming & Interactive
New Technologies
Production & Animation
Research & Education
Animation
Artificial Intelligence/Machine Learning
Computer Vision
Ethics and Society
Games
Generative AI
Industry Insight
Pipeline Tools and Work
Visual Effects
Full Conference Supporter
Full Conference
Experience
Similar Presentations
