AI Glossary · Foundations

Pre-training vs. Post-training

Pre-training builds general capability on broad data; post-training (SFT, RLHF, DPO) shapes tone, safety and task behavior.

Related terms in Foundations

Alignment: The discipline of making an AI system's behavior match human intent, values and organizational policy — both at training time and at runtime.
Artificial Intelligence (AI): Software systems that perform tasks normally requiring human intelligence — perception, reasoning, language, decision-making — by learning patterns from data.
Artificial General Intelligence (AGI): A hypothetical AI capable of performing any intellectual task a human can, across domains. Not a shipping product category today; treat vendor claims with scrutiny.
Benchmark: A standardized task suite (e.g., MMLU, GSM8K, HELM) used to compare model quality on reasoning, coding, safety or domain knowledge.
Chain-of-Thought (CoT): A prompting or training pattern where a model produces intermediate reasoning steps before its final answer, improving accuracy on math, planning and multi-hop tasks.
Classification: A supervised learning task that assigns inputs to discrete categories (spam vs. not spam, churn risk tier). The workhorse of predictive AI.