Newsletter 10: No escaping Gen AI

Generative AI went super mainstream this week!

Meta put its AI engine Llama 3 into popular mobile apps like WhatsApp, Instagram and Facebook Messenger. The functionality is rolling out to a few countries like the US, Singapore, Australia, New Zealand and Canada.

With Meta AI, you can generate text and images on the go for free, and the Meta AI engine even incorporates Google search results.

Screenshot of the Meta AI chat interface (Image source: Meta)

This means that very soon, most people will have easy access to Gen AI on their phones.

Think of Meta’s ridiculous reach: FB has 3 billion users, Instagram has 2 billion and WhatsApp has 2 billion (Statista data). In contrast, the world’s most well-known Gen AI app ChatGPT is estimated to have 100 million monthly active users in Feb 2024.

Microsoft has been trying to popularize ChatGPT by putting the Copilot icon on the Windows taskbar and creating a Copilot key for new Windows laptops. However, I don’t talk about Copilot much because it doesn’t work well (yet).

Meta and Google are now combining forces to pool their massive consumer user bases and technologies to knock Microsoft + Open AI off the podium. You can also use Meta AI on its own website.

And in a few months’ time, I’m sure Apple will join in the fray. Fun times!

All that said, I will still default to using ChatGPT 4 and Midjourney on my PC, not just because I pay for them. Why?

1. Gen AI power users need a physical keyboard and a mouse.

With ChatGPT 4, I paste in a lot of text, type many prompts, and extract the different text I need. I also create customized chatbots (GPTs) in ChatGPT 4 that save me a lot of time in specific work tasks.

It’s impossible to work like this on your phone. That’s also why laptops have yet to be replaced by phones for real work.

2. ChatGPT 4, Google Gemini and Meta AI churn out a generic style of images which I’ve come to dislike.

I use the free image engines in my Gen AI workshops to teach people how to create AI Art, but I always use the far-superior Midjourney for my own work. The Midjourney engine has a better sense of aesthetics, can be upscaled to high resolutions and have specific aspect ratios.

“Batman with violin”, generated in WhatsApp by Meta AI.
The same prompt, but in Midjourney.

3. It’s great to have Google search results in WhatsApp, but this is compromised by the risk of AI hallucination.

For example, I asked Meta AI in WhatsApp “Who is Ian YH Tan?” I never get the same answer. Sometimes Meta AI says it has no idea, sometimes it claims I’m a poet (I’m not), and sometimes it pulls the right content from my ThinkTan website or my media commentaries.

Part of the problem is that “Ian Tan” is such a common name today and I do want to change my name to something like Automa Tan…

While it’s great to see Gen AI going mainstream, I hope people will try out the paid ChatGPT for a month to see what the tech can truly do.

The mobile experience is always going to crippled by the screen size and user experience.

I wrote a lot on LinkedIn this past week:

So, you want companies to offer flexi-work…

This was my most popular post of the week but I nearly didn’t write it.

I am wary of writing about the topic of “flexi-work” because it is so divisive. IMO, it’s critical to understand how businesses work before workers demand a change in the working culture.

Why Jedi writers need not fear AI

Your resting heart rate always tells the truth about your body

How to save 30% on a Garmin that measures your heart rate

My lighting setup is finally done right

Early tests with Meta AI

That’s all, thanks for reading!