The AI app growing faster than ChatGPT 🚀

Hi Non-Techies,

I feel like AI has been in the news a lot this week, but it’s hard to tell because I’m surrounded by it all the time. Perhaps there are gardeners out there who feel as though the whole world is talking about whether their Chrysanthemums should’ve flowered by now.

But then I’ll stumble upon an incredible graph like this:

And I’ll think, “Nope, this isn’t just me. The world really is talking about AI all the time.”

Nowhere is that clearer than in my AI Academy. If you want to join a community of AI-curious Non-Techies from all kinds of professional and personal backgrounds, click the button below to start your 7-day free trial:

The AI app to rival TikTok 📈

Some of you might know that when the ChatGPT app was released, it reached 1,000,000 downloads faster than any other app in history.

Well, that record has just been broken. Thankfully for OpenAI, it’s been broken by another OpenAI product, Sora.

Sora 2, launched inside the Sora app this month, is a bit like TikTok, but every video on there is AI-generated. Users are encouraged to create their own 10-second AI videos using prompts, and can feature themselves, their friends (with permission) or some publicly available figures (like OpenAI’s CEO, Sam Altman - here’s a Sora video of him eating a cloud).

As you might imagine, the response has been mixed. Like so many AI releases, Sora 2 is startlingly impressive, scary and, at times, problematic, all at the same time.

Issues range from copyright infringement (the app is filled with intellectual property from films, TV shows and games) to Robin Williams’ daughter having to ask publicly not to have AI-generated videos of the late actor sent to her.

For me, there’s a lot of creative, funny and occasionally inspiring content coming out of Sora 2, but I’m also mindful of how easily it could be weaponised for online abuse. OpenAI will have to adapt quickly to curtail any misuse of such a powerful and popular tool.

It’s only available in the U.S. at the moment, with no clear timeline for a UK/EU release.

Are you rude to your AI tools? 😳

I tend to use “please” and “thank you” when I’m talking to my AI tools, especially if the tool is Claude. I find that a lot of AI consultants (certainly the ones that we work with at AIFNT, anyway) are in the same boat.

I’d probably need a separate newsletter to explain why, but if I had to guess, I’d say it’s 49% manners, 50% habit and 1% fear of a Terminator-style reckoning at some point in the future.

But a recent study has shown that rude prompts produce better results than polite ones. From the report:

“Contrary to expectations, impolite prompts consistently outperformed polite ones, with accuracy ranging from 80.8% for Very Polite prompts to 84.8% for Very Rude prompts.”

Here’s an example of a rude, “Man Men-style” copywriting prompt (possibly not safe for work, depending on your work).

How to tell an AI image from a real one 👀

A few newsletters ago, I introduced you to Nano Banana, Google’s scarily powerful AI image generator.

It’s so good, in fact, that the question of image authenticity - i.e. what’s real and what’s AI - has never felt more important.

Some smart Redditors may have discovered an answer. They’ve found that if you oversaturate an image generated in Nano Banana, it reveals a watermark. Here it is:

It reminds me of those optical illusions where, if you blur your eyes, you can see the outline of an elephant or something (I tried and, sadly, no elephant was found).

This is a really important topic. I want to live in a world where AI-generated content is easy to identify, not just for my own sanity, but because there are all kinds of dangerous implications if the line between real and AI content gets too blurred.

Whether or not these watermarks are the solution is another question entirely (presumably, it’s possible to remove them?), but for now, it’s comforting to know they exist.

A game-changer for marketers 🤝

I stumbled upon this clever use of LLMs (Large Language Models - like ChatGPT) the other week and thought, “I know someone who will find this interesting”.

The idea is to create a “synthetic consumer” to better predict (with 90% accuracy!) whether or not a customer will buy a product.

It’s from a paper titled “LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings”. Pardon? In Non-Techie speak, here’s how it works:

1) We give ChatGPT a demographic profile of a customer.

2) We give this “customer” a product and ask it for its thoughts on this product.

3) We then use another AI to rate the thoughts of this “customer” based on their probability of buying.

This is pretty cool. Imagine we want to launch a new product or service for our customers. Using AI, we can effectively create a testing group to see whether it’s viable, without actually having to launch the thing. Companies spend BILLIONS on this stuff.

If you’re feeling extra Techie, you can read the full paper (Warning: it’s very academic):

Some more interesting stuff we found this week:

🧠 This 10-minute tutorial video on how to combine NotebookLM with Perplexity for extra usefulness is really good.

🗑️ You can now permanently delete your ChatGPT chat history. ChatGPT conversations were being preserved as evidence in a lawsuit with the New York Times, but a judge has ruled they can be permanently deleted once more.

Okay, I’m going to spend the next week trying to muster up some rudeness to use in my prompts.

Starting from now - bye. Go away. Whatever.

(I just can’t - thanks for reading and please come back next week xx)

Heather and the AIFNT team.