I read the world’s most terrifying AI book — and it actually made me better at prompting

I use AI every single day. It's literally my job to test it, review it and try to break it. But after reading "If Anyone Builds It, Everyone Dies," a book about superhuman AI by Eliezer Yudkowsky and Nate Soares, I started thinking about AI differently. A lot differently.

Based on the title of the book it would be easy to think about AI in an overly dramatic, sci-fi, end-of-the-world way. But, surprisingly, the book didn't scare me into doomsday thinking. Instead, it made me more aware of using tools like ChatGPT, Claude and Gemini in a practical way. Essentially, the book changed how I use and experiment with popular AI tools.

The book’s central argument is about what happens if AI becomes too powerful to control. I’m not saying I walked away convinced every worst-case scenario will come true. It's a heavy book that you have to be ready for — I found myself reading it in smaller segements towards the middle because it was so heavy.

But one idea from the book really stuck with me: alignment. And once I understood that, I realized why AI sometimes gives answers that sound polished but still miss the point.

The realization that changed how I prompt

There are several good takeaways from the book. Many of which can be found on the accompanying website, but one of the more useful "everyday" takeaways for me was the idea that AI doesn’t need to be evil to be dangerous — it just needs to pursue the wrong goal really, really well.

That same concept shows up in everyday AI use. When ChatGPT gives you a generic response, misses your intent or confidently says something slightly off, it’s usually not because the tool is “broken.” It’s because it’s following the prompt too literally — or filling in gaps you never meant to leave open.

So now, instead of being casual about it, I started being much more deliberate. And the quality of my results improved almost immediately.

Here are a few examples:

• I stopped outsourcing the whole task. I used to hand everything over with prompts like, “Write a plan for this.” or "Help me make a decision" or "What should I do based on this information?" Now, I give AI the structure first. I decide the goal, the format and what matters most, then use AI to build from there. That one shift made my outputs feel less generic and much more useful.
• I started treating prompts like instructions. AI doesn’t naturally understand what you meant to say. It understands what you actually said. It's easy to fall into a trap that there is a thinking brain inside AI, but that's not true. It's getting better at reading patterns, but it's still very much artificial intelligence. So now I’m much more specific. I spell out what I want, what I don’t want and what a good answer should actually do. It feels a little excessive at first, but it cuts down on vague, robotic responses fast.
• I stopped trusting answers so quickly. One of the easiest traps with AI is mistaking confidence for accuracy. If something sounds a little too smooth, I slow it down. I ask it to explain its reasoning, show its steps or rethink the answer from another angle. That habit alone has saved me from using ideas that sounded smart but didn’t really hold up.

The takeaway

You don’t have to believe AI is going to destroy humanity to take one useful lesson from If Anyone Builds It, Everyone Dies. Tools like ChatGPT are only as good as the goals and instructions you give them.

That was the mindset shift for me. I’m far more intentional about how I use it and the results are noticeably better.

Have you read the book? Let me know in the comments what you think? I'd love to know what takeaways you've pulled from this important read.

Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.

More from Tom's Guide

• I asked ChatGPT to help me fix 7 everyday problems — I didn’t expect this
• I thought AI was getting smarter — until I saw the code that proves it’s ‘razzmatazzing’ us
• I kept hearing about a new TSA 'clear bag rule' — so I asked ChatGPT to get to the bottom of it