What AI did with 902 pages of car manuals

A modern car comes with an extraordinary amount of information. I discovered exactly how much when Adele recently bought her new Toyota Yaris Hybrid. The owner's manual is 507 pages. The navigation manual adds another 281. There's a 100-page multimedia guide and a further 14 pages of specifications for the car.

That's 902 pages altogether.

Somewhere amongst those 902 pages is probably the answer to almost any question we might have about the car.

Finding it is another matter.

And that presented exactly the sort of problem that makes me want to experiment with AI.

Rather than repeatedly searching through several manuals, I wondered what would happen if I gave ChatGPT all the official documentation and asked it to help me turn that mountain of information into something specifically designed for our particular car and the way we actually use it.

That's an important distinction.

I didn't want AI to tell me how a Toyota Yaris works based purely on whatever information it happened to have learned. I already had the source material. What I wanted AI to do was help me interrogate it.

Our situation is also slightly unusual because this isn't simply a family car. It's a driving school car. Adele will drive it herself, but numerous learner drivers will also use it every week. I'll occasionally drive it too.

We therefore started asking questions that probably aren't covered by a convenient section in Toyota's manual entitled "How to configure your Yaris when you're a driving instructor".

• How should the user profiles be organised?

• Should a particular profile be linked to the regularly used key?

• How can Adele quickly move between her own settings and the neutral settings used by pupils?

• What should happen with Apple CarPlay?

• Which settings should pupils be able to change?

  
  

Suddenly the exercise wasn't about summarising 902 pages of manuals. It was about taking reliable information and applying it to a real situation.

We eventually settled on an approach where the default profile could be used for pupils, with separate personalised profiles for Adele and me. We deliberately decided not to link the key regularly used with the car to Adele's profile because that could make switching between instructor and pupil use more awkward.

Then the experiment grew.

If AI could help us extract and organise the information, why not create a much simpler quick-reference handbook specifically for Adele's UK Excel Hybrid?

That's when I began to realise that the handbook itself might only be the beginning. Because the same principle could be applied to information for Adele's learners. Imagine a pupil struggling to understand one particular feature of the car. Rather than pointing them to page 300-and-something in a manual, we could create a simple one-page explanation specifically designed for a learner driver.

Text, diagrams and illustrations could explain what a control does, when it is used and what the pupil needs to remember.

Then take the idea another step.

A learner moving from a manual car to an automatic has a different set of questions.

• What changes?

• What habits from driving a manual do they need to stop?

• How should they use the accelerator and brake?

• What does the selector actually do?

• What mistakes commonly catch people out?

  
  

Again, there is an opportunity to take reliable information and turn it into something designed around the person who actually needs to learn it. At the moment I'm mainly thinking about written guides, diagrams and handouts. But why stop there?

AI-generated video is developing incredibly quickly. Could we eventually create short videos that demonstrate a particular driving skill or explain a car feature? Perhaps a pupil could watch a two-minute explanation before a lesson, then practise the same skill with Adele in the real car.

That becomes much more interesting than simply asking ChatGPT questions. It becomes a way to create a small library of learning materials specifically designed around Adele's pupils, her teaching, and the actual car they are learning to drive.

There is an important qualification to all of this.

Driving instruction isn't somewhere I'd want information invented by AI. That's precisely why the original source material matters. There is a big difference between asking AI, "Tell me how this works", and saying, "Here is the manufacturer's official documentation. Help me find the relevant information, check it against the source and turn it into something easier to understand."

The same principle applies to driving skills. The source needs to be authoritative, and the finished material still needs human checking.

AI doesn't remove that responsibility.

What it potentially removes is an enormous amount of the tedious work between finding reliable information and turning it into something useful. And that's the bit of AI that continues to fascinate me.

The clever part isn't necessarily asking a clever question. Sometimes it's giving AI good source material, explaining the real-world problem properly and then working with it to produce something that didn't previously exist.

Our experiment started with 902 pages of Toyota manuals and one simple question:

Can we turn all this into something Adele can actually use? We can.

But the more interesting discovery is that the answer opened the door to another question.

What else could we create?

Handbooks. Quick-reference sheets. Learner handouts. Diagrams. Explanations of driving techniques. Perhaps eventually short training videos.

Suddenly those 902 pages don't look like an enormous pile of documentation. They look like source material.

And that is a very different way of looking at information, and why I am so interested in AI as a 'tool' to make our jobs easier and more creative than ever before.