Temperature, guardrails, few-shot, hallucination and MCP — quick, precise definitions for the words getting thrown around in every planning meeting.

Every team has that meeting where someone says a term with total confidence and three other people nod without actually knowing what it means. Here are five that come up constantly right now, defined simply enough to say out loud in your next standup.

1. Temperature

Temperature controls how “creative” or random a model’s word choices are. Low temperature (near 0) makes it pick the most likely next word almost every time, giving consistent, predictable answers — good for factual tasks.

High temperature makes it take more risks with word choice, giving more varied, surprising output — good for brainstorming, riskier for facts.

2. Hallucination

This is when a model states something confidently that simply isn’t true — a fake citation, a function that doesn’t exist, a statistic it made up.

It happens because the model is built to produce plausible-sounding text, not to look facts up, so “sounds right” and “is right” aren’t the same thing to it. The fix isn’t asking it to “not hallucinate” — it’s giving it real information to work from (see RAG, above) and checking its claims.

3. Few-shot prompting

Instead of just asking a model to do something, few-shot prompting means showing it two or three examples of the exact input-and-output pattern you want first.

It’s the difference between telling someone “format this nicely” and showing them one finished example — the second gets you a far more consistent result.

4. Guardrails

Guardrails are the checks and limits placed around a model’s input and output: blocking certain requests, filtering unsafe responses, or validating that output matches an expected format before it’s used.

Think of them as the seatbelt, not the driver — they don’t make the model smarter, they limit the damage when it gets something wrong.

5. MCP (Model Context Protocol)

MCP is an open standard that lets an AI model connect to external tools and data sources — your files, a database, a project-management app — through one consistent interface, instead of every app needing a custom one-off integration.

If you’ve seen an AI assistant read your calendar or search your company’s docs, there’s a decent chance MCP, or something like it, is the plumbing making that connection possible.