Results for "safety cost"

Cost to run models in production.

Cost of model training.

Assigning AI costs to business units.

Optimal pathfinding algorithm.

Tradeoff between safety and performance.

Accelerating safety relative to capabilities.

Automated detection/prevention of disallowed outputs (toxicity, self-harm, illegal instruction, etc.).

Systems where failure causes physical harm.

Finding control policies minimizing cumulative cost.

Optimal control for linear systems with quadratic cost.

Mechanism to disable AI system.

Hard constraints preventing unsafe actions.

Techniques to handle longer documents without quadratic cost.

Observing model inputs/outputs, latency, cost, and quality over time to catch regressions and drift.

Limiting inference usage.

Methods for breaking goals into steps; can be classical (A*, STRIPS) or LLM-driven with tool calls.

Optimizes future actions using a model of dynamics.

Optimizing continuous action sequences.

Mathematical guarantees of system behavior.

Sudden jump to superintelligence.

Risk threatening humanity’s survival.

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

Of predicted positives, the fraction that are truly positive; sensitive to false positives.

Of true positives, the fraction correctly identified; sensitive to false negatives.

Of true negatives, the fraction correctly identified.

Selecting the most informative samples to label (e.g., uncertainty sampling) to reduce labeling cost.

When some classes are rare, requiring reweighting, resampling, or specialized metrics.

How many requests or tokens can be processed per unit time; affects scalability and cost.

Optimization using curvature information; often expensive at scale.

Routes inputs to subsets of parameters for scalable capacity.