Uncle’s Diary · Learn AI Simply · 5 minutes · No tools required
I’ve spent two entries — How to Test an AI Tool Without Spending a Dime and I Didn’t Even Know It Was Broken — telling you how little of this I actually do myself.
This one’s different. This one was all mine, and it took about four minutes.
What I did
I had a voice for my dog and I couldn’t decide if it was right. Not whether it was good — I’d already been told it was technically fine, the pauses trimmed, the timing sorted. I mean whether it was him. Whether that was the voice of the animal currently asleep on my foot.
There is no tool for that. There is no prompt for that. I checked.
So I sent it to three people and asked what they thought. Here is everything I got back, word for word:
“The voice with the British accent is perfect!”
“Boomer loves it!”
“I love it too!”
Three for three. No hesitation, no “but”, nobody asking to hear the other version.
So it’s decided. That’s the voice. It’s not changing.
🐾 Boomer here. I should like the record to show that the accent was never in question. I am a King Charles Cavalier Spaniel. The accent is not a stylistic choice, it is a birthright. What is notable is that Uncle needed three other people to tell him something I have been communicating by expression alone for five years. Still. He asked. Credit where it’s due, and it is rarely due.
The honest ledger, for once in my favour
I’ve been running this ledger every time to show you how much gets done for me. Here’s what it looks like now.
What I did:
Decided the question mattered enough to ask. Worked out that it wasn’t a question a machine could answer. Chose who to ask. Sent it. Read the answers. Made the call and closed the decision.
What was done for me:
Nothing. Not one thing.
Twice running, that box was lopsided the other way and I published it anyway. So I’m allowed to point out that this time the long list is mine.
And notice what kind of work it was. Not technical. Not clever. I didn’t learn anything to do it. I recognised that a question had left the machine’s territory and entered the world where people live, and I walked it over there.
That’s a skill. It’s just not the sort anyone gives you a certificate for.
Now the uncomfortable part
Three people said yes and I’m going with it — and I want to be straight with you about what that yes is worth.
Those three people love me. They were never going to tell me my dog sounded wrong. Not because they’d lie — they wouldn’t — but because they were listening to it the way you listen to something a person you love has made. Which is warmly. Which is exactly what you want from your family and exactly the wrong instrument for measuring a stranger’s reaction.
So what did I actually learn?
I learned the voice isn’t bad. That’s real information and I’ll take it. If something had been properly off — the accent grating, the pace odd, the whole thing landing as creepy rather than charming — I’d have heard about it. Affection doesn’t paper over a genuine wince. Three people wincing politely is a signal you can read from space.
What I did not learn is whether it’s good. Good means it works on someone who owes me nothing. My family can’t tell me that, because they aren’t strangers and can’t pretend to be.
Ask the people who love you whether it’s wrong.
Ask strangers whether it’s right.
Both questions are worth asking. They’re just different questions, and the mistake — the one I nearly made about an hour ago — is thinking the first one answers the second.
Why I decided anyway
Here’s where I’d push back on my own caution.
I could have gone looking for better evidence. Posted it somewhere, asked a forum, run some sort of poll. I’ve done that sort of thing before and I know exactly where it ends: three weeks later, forty opinions, no decision, and the project quietly dead in a folder.
The voice was blocking everything behind it. Episode one. The talking bit on this website. All of it waiting on a decision I kept describing as “not quite settled.”
So I took the weak evidence and I closed the question. Deliberately, knowing it was weak, writing down here that it was weak so nobody can accuse me later of pretending otherwise.
A decision made on thin evidence beats a decision deferred on no evidence. You can revisit a choice. You cannot revisit the six months you spent not making one.
And it’s reversible. If strangers turn out to hate it, I change the voice. It’s a setting. It’s not a tattoo.
Your turn
You’ve probably got something sitting unfinished right now because you can’t tell whether it’s any good.
Don’t ask an AI. It’ll tell you it’s great — that’s very nearly its default setting and you should discount it accordingly.
Send it to three people instead. And ask better than “what do you think,” because “what do you think” gets you “nice!” and “nice!” is worth nothing. Ask this:
“Don’t tell me if you like it. Tell me if anything about it makes you wince, or feels off, or would make you stop watching. If nothing does, just say ‘nothing’ and I’ll take that as the answer.”
Giving people permission to say “nothing” is what makes it usable. It stops them hunting for a compliment.
Then decide. Today, not eventually.
What’s actually next
The voice is locked, so everything behind it comes unstuck. Which means the next real test isn’t three people who love me.
It’s strangers, on the internet, who don’t.
That one I can’t control, can’t prompt my way around, and can’t ask anyone to handle on my behalf. I’ll tell you honestly how it goes.
For the record. Three responses, quoted above in full. I haven’t trimmed anything unflattering, because there wasn’t anything unflattering — which is the whole point of the middle section of this one.
Next: the first episode. Made, published, and judged by people who have no reason to be kind.
The voice was made with an AI voice tool in July 2026. The three responses were collected the same month. Your own family will be similarly generous — plan for it.