Alignment
Alignment is the work of making an AI system do what you actually meant, not merely what you literally typed: its goals, limits and refusals are shaped during training and deployment so its behaviour stays within what operators and society intend.
No jargon, no math
In plain words, what is it?
A model trained to predict the next token knows nothing about your intent; it optimises for a plausible answer. Alignment is the stack of techniques that bends that optimisation toward your real goal. It starts in training: after the base model learns language, human raters score its answers and a reinforcement pass rewards the helpful, honest, harmless ones (that is RLHF). It continues at deployment: the system prompt sets the role, safety fine-tuning teaches refusals, and guardrails bound what the model may do or say. You see alignment working when a model declines to invent a discount or flags an ambiguous order instead of guessing. Its limit: alignment is statistical, not absolute. Unusual phrasing can slip past it (that is a jailbreak), which is why aligned models still sit inside guardrails and audits.
Terms that live nearby.
RLHF
RLHF, reinforcement learning from human feedback, is the technique that tunes a language model's behaviour with human taste: people rank sample answers, a reward model learns to predict those preferences, and the language model is trained to score higher.
ReadFine-tuning
Fine-tuning is additional training that teaches a general model your specific patterns: you show it hundreds or thousands of examples of your tasks, your tone and your formats, and it adjusts its internal weights until that behaviour becomes its default.
ReadGuardrails
Guardrails are the checks and limits wrapped around an AI model so its behaviour stays inside your rules: they screen what goes in, constrain what the model can do, and verify what comes out before it reaches a person or a system.
Read§ 03
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.