Chapter 06 · 15 minutes
First chat and real scenarios
Turn off the model's thinking out loud, ask useful questions, and learn where a small local model shines and where it falls over.
Goal
Have a real conversation, try it on everyday scenarios, and see its limits with your own eyes.
One setting first: Reasoning off
Qwen3.5 models "think out loud" before answering by default. Great for puzzles, slow for chat, especially without a graphics card. Jan has a Reasoning control with auto, on and off. Set it to Off for this chapter.
Source: Qwen3.5 model card (thinking on by default) and Jan v0.8.8 (reasoning control).
Your first chat
- Click New Chat.
- Pick Qwen3.5-2B in the model picker.
- Set Reasoning to Off.
- Ask: Explain what a CPU does, like I am 12.
The first answer is slower because the model is loading into memory. Then it speeds up.
Scenarios to try
| Who | Try this |
|---|---|
| Student | Quiz me with 5 questions on photosynthesis. Wait for my answer each time. |
| Job hunter | I have an interview for a data analyst internship tomorrow. Give me 5 likely questions and a short tip for each. |
| Office worker | Rewrite this email to sound polite but firm: (paste a draft) |
| Small business | Write a WhatsApp message announcing our Diwali offer, under 50 words. |
| Developer | Explain this error and suggest a fix: (paste an error message) |
| Privacy-minded | your three things from chapter 01 |
Where it falls over
Try these on purpose, so you know:
- Who won the most recent cricket World Cup? It does not know recent events, and may answer confidently anyway.
- What is 4,837 × 2,961? Small models are bad at exact arithmetic. Use a calculator.
- Ask the same factual question twice in new chats. The answers may differ.
Rule of thumb: treat it like a smart friend, not like a doctor, lawyer or search engine.
See your laptop working
While an answer is being written:
- Windows: Task Manager → Performance. CPU or GPU jumps.
- Mac: Activity Monitor → CPU or GPU history (Window → GPU History).
That is your laptop doing what a data centre usually does.
Check it works
You got sensible answers to at least three scenarios, and you saw one wrong or made-up answer.
Exercise
Turn Reasoning to On and ask the arithmetic question again. Compare the wait and the answer with Reasoning off. When is the wait worth it?