Results for "objective design"

A scalar measure optimized during training, typically expected loss over data, sometimes with regularization terms.

Visualization of optimization landscape.

Designing systems where rational agents behave as desired.

Designing efficient marketplaces.

Ensuring AI systems pursue intended human goals.

Average of squared residuals; common regression objective.

Training objective where the model predicts the next token given previous tokens (causal modeling).

Optimization under uncertainty.

Goals useful regardless of final objective.

Converts constrained problem to unconstrained form.

The text (and possibly other modalities) given to an LLM to condition its output behavior.

System design where humans validate or guide model outputs, especially for high-stakes decisions.

Tendency to trust automated suggestions even when incorrect; mitigated by UI design, training, and checks.

Mathematical framework for controlling dynamic systems.

Robots made of flexible materials.

Competition arises without explicit design.

System-level design for general intelligence.

Learning a function from input-output pairs (labeled data), optimizing performance on predicting outputs for unseen inputs.

Learning where data arrives sequentially and the model updates continuously, often under changing distributions.

A parameterized mapping from inputs to outputs; includes architecture + learned parameters.

Automatically learning useful internal features (latent variables) that capture salient structure for downstream tasks.

Configuration choices not learned directly (or not typically learned) that govern training or architecture.

A function measuring prediction error (and sometimes calibration), guiding gradient-based optimization.

Predicts masked tokens in a sequence, enabling bidirectional context; often used for embeddings rather than generation.

A gradient method using random minibatches for efficient training on large datasets.

Constraining outputs to retrieved or provided sources, often with citation, to improve factual reliability.

A model that assigns probabilities to sequences of tokens; often trained by next-token prediction.

A preference-based training method optimizing policies directly from pairwise comparisons without explicit RL loops.

Ensuring model behavior matches human goals, norms, and constraints, including reducing harmful or deceptive outputs.

Maliciously inserting or altering training data to implant backdoors or degrade performance.