In this week’s newsletter, I wrote an essay about how I am reworking my workshop definition of Gen AI. I recap my media interviews with CNA in the past week, the sudden launch and even more sudden shutdown of Claude Fable 5, explain why AI visuals are making you nauseous, and other non-AI stuff like health, nature, art.

Greetings, Earthlings! I was thankful to be interviewed by Channelnewsasia twice this week (live and video podcast), and I’ve written a short essay below on how I’m redefining AI as a “data processor” when I teach it to people.

Bishan Stadium, 6th June 2026

AI Stuff

CNA Deep Dive interview – Are educators, fresh graduates and employers aligned on AI?

Here’s the shorter video reel: Avoid hiring managers who ask for AI literacy without being AI literate themselves.

CNA Talkback with Daniel Martin: How do you know your news wasn’t written by AI?

Claude Fable 5 was launched to much excitement, then taken off the shelf after the US Govt banned foreigners from using it. To be honest, I only played with it for a while, I was too busy meeting up with people this week

If you feel nauseous looking at AI images, it is because AI engines have no intrinsic sense of aesthetics or taste.

Non-AI writings

This post about my obituary photo must have freaked people out because very few dared to comment.

I’ve been buying different gadgets to study my body statistics. I can recommend this Omron body composition monitor.

Photos of trees from Bishan Park.

Practicing my Chinese handwriting. It’s actually pen calligraphy and takes years to train.

I jogged 9km just to buy soya bean milk.

Applying Michael Pollan’s eating advice.

Sunday Essay

After three years of coaching people in Gen AI and many discussions with intelligent friends, I am finetuning how I define the technology to others.

Previously, I told people: “Gen AI is a tool, not a miracle worker. You have to call the shots!”

Now, I will tell people: “Gen AI is not an answer machine. It is a probabilistic data-processing tool. You are responsible for the data you give it, the output it produces, and what you do with that output.”

ChatGPT provided an easier version: “Gen AI is not a truth machine. It is a very good guess-and-mix machine for data. It takes your input, combines it with patterns it has learned, and produces output that may be useful. But you must always check it before you use it.”

The new spiels are more technical and not as easy to remember. However, Gen AI is a technical thing and people need to get more technical about it. I’m hoping the concept of data processing will get users to realise that they need to be data literate first.

This is the data-driven thinking process people need to have when using any technology to solve problems. It doesn’t matter if you are using Excel or ChatGPT, you must ask these questions:

What is the problem I need to solve?
What data do I need?
What data is out there? How do I find it?
Are there multiple sources of data for me to triangulate it?
Is this data accurate, reliable, objective, and verifiable?
Does the data need cleaning and filtering?
What knowledge or insights can I get from this data?
What other data do I combine this data with?
What is the data output that I want, after processing?
How do I adapt the data output to different forms?
Is the output data accurate and valuable?
Will I take responsibility for distributing the data?

The above steps are complicated, but any responsible and skilled worker should already have these steps largely burned into their psyche. I learned all these steps as a journalist nearly 30 years ago, but I’ve never had to describe it like this until Gen AI came along to expose our workflows and thinking processes.

I strongly believe that if you are not strong with managing data (data literate), then you will find it difficult to wield AI well. This supports my first principle that you need to be good at your job to wield AI well. If you are good at your job, then you know what data you need to solve the job-specific problems that come your way.

For example, a good car mechanic would have plenty of knowledge in his head to assess why a car component is not working. He would know which tools to use, how to use those tools, and how to evaluate if his repair work was done well. If he doesn’t have enough info (eg. how the car got damaged), he would ask the car owner specific questions. If the car was not repairable, he would let the car owner know. If the mechanic were to use any AI tool for assessment or repair, he would provide relevant data and evaluate the outcome, and be accountable to the customer.

Problem-solving relies on getting enough data, recombining data, and taking action on the data. If you want AI to be useful to you, be good at using data.