Welcome back to Retire & RISE {{first_name}}
Two weeks ago I asked ChatGPT to tidy up a paragraph I'd written for this newsletter. It came back better than what I gave it. Tighter, same meaning, still sounded like me. Ten minutes later, in the same conversation, I asked what time the hardware store on the square closes on a Saturday. It told me. Specific time, no hedging, sounded exactly as sure of itself as it had about the paragraph. It was wrong, and the store had been shut an hour by the time I got there.
By the end of this you'll:
Know why the same tool can be excellent and useless ten minutes apart, and why that isn't something you did wrong
Have a two-minute way to find out what ChatGPT is steady on and where it wanders
Learn the difference between an answer that's steady and an answer that's right, which are not the same thing
Understand which jobs are safe to hand it and which ones really aren't
🧠 It's Uneven, Not Unreliable
Here's the thing nobody tells you at the start. ChatGPT isn't uniformly good or uniformly bad. It's very strong at some jobs and not good at others, and from the outside those two kinds of job look identical.
MIT Technology Review described a study that gets at this nicely. Researchers built an AI like the one behind ChatGPT and trained it on a huge set of New York taxi journeys. It could give good directions from one point in Manhattan to another. Then they forced it to take a detour. It failed completely. Not "got a bit worse" – failed. It had learned the routes without ever learning the city.
That's the shape of the thing. Fine on the ground it knows, lost the moment you step off it.
And here's the part that actually matters: it sounds the same either way. My paragraph and my hardware store got the same confident tone. There was nothing in the answer itself telling me one was solid and the other was invented. The errors aren't the real problem. The even voice is.
🔁 Ask It Twice
So don't judge it by how sure it sounds. Test it, on a question you actually care about. Pick something you already know a bit about – not to catch it out, but so you can judge the answers when they come back.
→ Take one real question, ask ChatGPT, then ask the same thing in different words in a second, separate chat. Put the two answers side by side.
Three things worth knowing before you try it:
Make the second one a Temporary Chat. This matters more than it sounds. ChatGPT remembers – within a conversation, and these days often across your other chats too – so an ordinary new chat may just repeat what it already told you. A Temporary Chat doesn't use or save any of that. It's a toggle near the top of the chat screen.
Change the wording, not the question. "What are the best ways to cut a grocery bill?" and "How can I spend less at the supermarket each week?" are the same question dressed differently.
Read for agreement, not for quality. Where the two answers say the same thing, you're on ground it knows well. Where they part company, you've found an edge.
This isn't my invention. Consumer Reports put the same safety questions to five different chatbots and graded the answers against their own experts. The results ran from perfect to badly wrong, and one of the habits they recommend is exactly this – ask more than one way and compare. The other worth stealing: ask it where it got that. Their write-up is here.
🗺️ Steady Is Not the Same as Right
Two matching answers tell you the tool is on familiar ground. They do not tell you the ground is true. It will give you the same wrong closing time twice, with the same steady voice, exactly as it gave me one. So when the two agree, that has earned the answer a look. It hasn't earned your trust.
What you're building is a map of your own, not a general opinion about AI. Rewriting a letter, steady. Explaining a term you half-know, steady. Opening hours, prices, dates, anything about a specific real thing out in the world, go and check it every time, however sure it sounds.
That map is personal. Mine won't be yours, because we ask different questions.
Skip the whole exercise and you land in one of two places. Either you believe it because it sounds sure, which is how I ended up at a closed hardware store on a Saturday. Or you decide the thing can't be trusted at all and leave a genuinely useful tool on the shelf. The middle ground is the good place to be, and the only way in is testing.
📣 Something I Looked Into
A second word on the training I mentioned Tuesday. The most sensible objection to it is one I had myself: it doesn't run until October, so why decide now?
You don't wait for October. Buying puts you into the Challenge Hub, their members' area, the same day. The setup steps, the pre-challenge material and the Affiliate DNA Quiz – a questionnaire that points you at one approach, not five – are there immediately, not held back until October 6th. The three live afternoons with John and Alex Thornhill are the event, but the material starts when you do.
So, plainly, what $9 buys: three live afternoons, the workbook, the quiz, the private community, and 72 hours of replays. Their page says $9 holds until September 14th and more than doubles after that. There's a $47 All-Access upgrade offered on top at checkout, and smaller things offered later.
Skip it if you need money in the next few weeks, or if three afternoons in October aren't realistic. This is not fast money and it is not a light commitment. I'd rather you knew that now.
I earn a commission if you join through my link, and more if you take the upgrades I tell you to be careful about. It costs you nothing extra – the price is the same either way.
Here's my honest write-up, which names every extra before you'd pay for any of it.
🧭 Before You Close This
Do it once, on a question that matters to you, and you'll know something about this tool that no article can hand you – including this one.
If you try it and the two answers surprise you, hit reply and tell me which question you used. Just the question is plenty.
Talk soon,
Bob
A short story from my place, about my dog Bradley and the day a new gadget showed up – no sales in it.
#RetireAndRise #AIMadeSimple #ChatGPT #OneSmallStep #Over50
🔁 Hashtags
#RetireAndRise #AIMadeSimple #ChatGPT #OneSmallStep #Over50 #SeeYouAtTheBank
