Results for "reliability"

Letting an LLM call external functions/APIs to fetch data, compute, or take actions, improving reliability.

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

Measure of consistency across labelers; low agreement indicates ambiguous tasks or poor guidelines.

Systems where failure causes physical harm.

A mismatch between training and deployment data distributions that can degrade model performance.

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

The degree to which predicted probabilities match true frequencies (e.g., 0.8 means ~80% correct).

A proper scoring rule measuring squared error of predicted probabilities for binary outcomes.

Architecture that retrieves relevant documents (e.g., from a vector DB) and conditions generation on them to reduce hallucinations.

Automated testing and deployment processes for models and data workflows, extending DevOps to ML artifacts.

Model-generated content that is fluent but unsupported by evidence or incorrect; mitigated by grounding and verification.

A broader capability to infer internal system state from telemetry, crucial for AI services and agents.

Systematic review of model/data processes to ensure performance, fairness, security, and policy compliance.

System for running consistent evaluations across tasks, versions, prompts, and model settings.

Stress-testing models for failures, vulnerabilities, policy violations, and harmful behaviors before release.

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

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.

Forcing predictable formats for downstream systems; reduces parsing errors and supports validation/guardrails.

Measures divergence between true and predicted probability distributions.

Systematic error introduced by simplifying assumptions in a learning algorithm.

Models evaluating and improving their own outputs.

Required human review for high-risk decisions.

Maintaining two environments for instant rollback.

Centralized repository for curated features.

Measure of spread around the mean.

Maintaining alignment under new conditions.

Sampling multiple outputs and selecting consensus.

Applying learned patterns incorrectly.

Model trained on its own outputs degrades quality.