A base model that just won't stop asking questions

I asked it a plain question, and it answered with three more questions of its own.

Beginner to intermediateRead 5 minDoing: about an hour6 steps

▶ Start

What you'll have at the end

Two saved replies to the same prompt, from a model's base build and its instruct build, correctly labelled

Before you start

You need

Access to an account or library that can call a specific open model checkpoint by name and print back its raw reply, plus enough scripting to send one request and read the result.

Not covered

Which exact model family or hosting platform to standardize on, or how to fine-tune a base checkpoint into a custom assistant.

Leans on

Large Language Models (LLMs) Cheat SheetIts pre-training and fine-tuning tables name the exact two stages that separate what you sent from what actually answered.
Foundation Models in AI Cheat SheetIts own instruction-tuning row backs the same fix with a second, independent account.

6 steps

Where to go next

Read the model card, then decide if you can trust it

for checking what a specific checkpoint was trained on and where it breaks, once you already know which build you're calling

Checked 18 Aug 2026

Part of the Large Language Models (LLMs) cookbook