What 11 People Taught Me About AI Training in Two Days

What 11 People Taught Me About AI Training in Two Days

Real numbers from a recent programme and the one honest caveat most trainers leave out.

Last week I stood in a training room in Senai with a group of technical professionals from a government-linked technical organisation. Two days later, the room had changed. Not because AI is magic, but because the way they used it had a method for the first time.

I want to show you the numbers, because I believe training should be measured, not just felt. And I want to show you the honest limits of those numbers, because that honesty is the whole point.

The results

We assessed every participant before the programme and again after, then matched the responses person by person. Eleven participants completed both. Every figure below was recomputed directly from the raw data, not estimated.

• Average knowledge score rose from 60.6 percent to 87.1 percent. That is a gain of 26.5 percentage points.
• Average confidence rose from 2.78 to 4.56 out of 5.
• Every question on the assessment improved. Every participant improved in knowledge, confidence, or both.

Where the real learning happened

The headline numbers are pleasant. The story underneath them is more useful.

The largest jumps were not on the exciting topics. They were on the disciplines that keep AI safe and reliable at work.

Writing a clear prompt using PIDO, which stands for Parameter, Instruction, Details, and Output, moved from 9 percent to 91 percent. Understanding how ChatGPT actually works- that it predicts language rather than knowing facts- moved from 46 percent to 100 percent. Knowing the right first step when AI gives a confident answer with no source, which is to verify it, climbed sharply too.

In other words, people did not just learn to type better prompts. They learned to think before they trust.

The honest caveat

Here is the part many trainers skip. These are immediate learning gains. They are strong evidence that people understood and felt more capable on the day. They are not, on their own, proof of lasting productivity or changed habits back at the desk.

Real behaviour change is decided in the 30 to 60 days after the room empties. It happens when one person takes one repeated task, builds one reusable prompt, checks the output, and measures whether it genuinely saved time or improved quality. That is why every programme I run ends with an application plan, not a round of applause.

If a training provider only shows you the after number, ask them what happened a month later. The answer tells you whether they trained people or just entertained them.

The philosophy behind the method

Everything in the programme sits on one idea I call Bits vs Bricks.

AI is excellent at the bits: drafting, summarising, reformatting, comparing, analysing. Humans must keep the bricks: the purpose, the judgement, the accountability and the final decision. A tool can help you build faster. It cannot decide what is worth building, or take responsibility when it matters.

I summarise it in four words at the end of every session:

AI Draft. Human Final.

And in one line for the teams I train in Malaysia and Brunei:

Do not only use AI. Create with it.

Work with me

I deliver AI and productivity programmes for government agencies, corporate teams, universities and community organisations. The Effective Use of ChatGPT & AI series takes a team from copying and pasting with no method to structured, safe, and genuinely useful AI habits, with the follow-up plan to make it stick.

If your team is busy and wants results you can defend with data, let us talk.

Start here: linktr.ee/coachamirul

Tags: AI training, ChatGPT, prompt engineering, PIDO, learning and development, productivity, Malaysia, future of work, HRD Corp