Concept

The ability of GenAI systems to consistently perform their intended functions across a range of technical conditions, which include adversarial inputs, noisy and incomplete input data and the inherent stochasticity in the generative process.

The resilience of GenAI systems in varied deployment environments and operational settings, including the capability of the systems of being efficiently and effectively fine-tuned or adapted to different tasks and domains without significant loss in performance and security.

The capability of GenAI systems to engage in meaningful and contextually relevant interactions with a diverse range of users, each with unique behaviours, cultural backgrounds, and prone to human errors. It also encompasses resilience to ethical, fairness, and misuse risks of GenAI.