Results for "trial-and-error"
Cost to run models in production.
Vector whose direction remains unchanged under linear transformation.
Matrix of first-order derivatives for vector-valued functions.
Belief before observing data.
Sampling from easier distribution with reweighting.
Minimum relative to nearby points.
Optimization under uncertainty.
Correctly specifying goals.
Maintaining alignment under new conditions.
Willingness of system to accept correction or shutdown.
Task instruction without examples.
Using markers to isolate context segments.
Sampling multiple outputs and selecting consensus.
Breaking tasks into sub-steps.
Computing end-effector position from joint angles.
Artificial environment for training/testing agents.
Directly optimizing control policies.
Space of all possible robot configurations.
Perceived actions an environment allows.
Physical form contributes to computation.
Learning without catastrophic forgetting.
Unequal performance across demographic groups.
Requirement to preserve relevant data.
Fabrication of cases or statutes by LLMs.
Requirement to reveal AI usage in legal decisions.
Credit models with interpretable logic.
No agent can improve without hurting another.
Truthful bidding is optimal strategy.
Competition arises without explicit design.
AI tacitly coordinating prices.