open-ended

Definition

Describes tasks or environments with no single correct answer and no natural termination — open-ended generation, open-ended exploration, agents that keep inventing their own goals. The defining difficulty is evaluation: with no ground-truth target, quality has to be judged by humans, by other models, or by proxy metrics, all of which are contestable. It stands opposite closed tasks like classification, where the answer set is finite and fixed.

Explainers using this term

• Paper Deep Dive — Large Discovery Models: giving an LLM a value signal for what to try nextLarge Discovery Models: Empirically-grounded Model-Based Open-Ended Search
• Paper Explained: JoyAI-Video-Edit — Autoregressive Diffusion That Keeps Editing 720p Video at ~30 FPS, Without Ever Seeing the FutureJoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion