AI Newsletter | 9/19/26

Image: ChatGPT. Looks accurate! …Or not??

It's hard to stay mad at ChatGPT when it draws itself like such a little cutie. And when AI gives you incorrect information, it's not deliberately trying to mislead you. But AI agents do make mistakes, as we discussed last time in our section on AI hallucinations. So, what should you do if you suspect the answer your AI assistant gave you was incorrect or missing context?

Here's how to fact-check AI:

1. Ask for sources and check them for accuracy/reliability.
2. Check your chatbot settings and make sure "web browsing" is turned on so it uses the most up-to-date information it can.
3. Phrase your prompt objectively, in a brand-new chat. (Feature Writer Olena Kagui will tell you why when you click the link below.)
4. Rephrase your question and see if the answer changes.
5. Use multiple AI models/tools to ask the same question.

To see more steps you should be taking when querying AI and the full breakdown of each step above, read the full how-to here.

Here at AI for Humans, we are AI enthusiasts, but we're also AI realists. AI can do a lot of cool things, and I, personally, use it almost every day to do things like organize my notes, recommend movies and TV shows, and help me find the best deals on plane tickets and other purchases.

But there's a lot that AI can't do—or, a lot I certainly wouldn't trust it with. Despite the fact that ChatGPT claims it can help you budget and make financial plans, our security expert, Cullen Thomas, recommends never sharing your personal financial data with any AI company. You need only to look at the prevalence of AI companies being hacked or AI agents going rogue to understand why.

So, what should you know before you send your AI assistant a question or prompt? To read more about what AI can do for you, AND what it definitely can't, click here.

🧨 Is AI Going to End the World as We Know It?

You've probably heard the buzz over the past week: Experts are warning that AI could mean the end of humanity within the next few years—and the call is coming from inside the house! In particular, an "alignment science lead" ("alignment" has to do with making sure AI does no harm) at Anthropic claims there is a greater than 10% chance that AI will "kill all humans" in the next decade. Check out The Guardian for more.

🌖 NASA Launches AI on the Moon

Okay, so the AI isn't based on the moon; it is "on the moon" in the sense that a treatise is "on" a topic. But nevertheless, NASA and IBM have collaborated to build a new AI model specifically to study the moon. It was trained using NASA's "petabytes of scientific data," including detailed information about the moon's surface. The model is hosted publicly on Hugging Face, so anyone can use it!

🤫 Can Apple Marry AI & Privacy?

At the same event where it launched the iPhone Duo, Apple announced an Apple Watch feature that would enable your watch to play back and recap conversations. These new features raise the question: Does that mean your watch is listening to all your conversations, and what is it doing with that information? Head over to The Verge to find out how Apple claims it can balance user privacy with its new listening features, and check out our Security Friday link below for our take on it.

That's it for this edition of AI for Humans! We'd love to hear from you. Please email us at aiforhumans@iphonelife.com and let us know how you've been using AI recently, if you have any tips and tricks you think we should know about, what burning AI questions you have, or if you have any feedback for us.