Day 2 of training my AI twin as a closed beta user

It’s day two of training my twin on Tsuin, and I have to be honest, it’s a tedious process. However, it also exposed my reasoning a lot.

I was talking to the wrong version

On Day 1 I said the twin starts in a pre-school phase and says “I don’t know” until you train enough examples. That was actually a bug.

The founders of tsuin.ai told me a bug had me using the untrained version of my twin. That’s why it couldn’t answer anything I had just trained. Now everything works, and it’s able to retain facts and reasoning to answer me.

Explaining my reasoning exposed the flaws

What used to feel natural to me fell apart once I said my reasoning out loud. I realised there are so many flaws in my reasoning, and the way I explain things is confusing.

The most frustrating part is that I need to clearly explain my reasoning to the twin. I hate to admit it, but all this time my way of explaining something was a mess.

So I use it to see where my reasoning should improve. I can see how I construct my sentences, my logic and how I explain something.

AI was doing my thinking for me

Honestly, I use AI a lot, and I noticed I’ve been offloading a lot of cognitive work to it. I would just not think about it, and that has been quite rotting my brain.

In a way, giving my reasoning to the twin is training my brain to think better.

I explain the why behind every correction

If I just correct the twin, it doesn’t pick up my reasoning. An AI twin is supposed to do tasks for me, so it has to act like me.

For example, when I use Claude or Codex and I see something is missing, I ask it to clarify with me. I deal with people the same way. When someone explains something and I don’t have enough context, I clarify with them to understand the situation.

I want my twin to have the same habit. When it doesn’t have the context, it should ask the AI agent to explain more, or come back and clarify with me.

My best version for my lowest moments

When I’m tired, I start to think in shortcuts. Sometimes I’m not aware of something I should have paid attention to, or a lesson I learned before.

A computer doesn’t forget as easily as I do, and it doesn’t get tired. So I’m putting as much knowledge and wisdom as I can into my twin, so it keeps the best version of me for the moments when I’m not at my best.

That’s why I don’t train it with questions I already know the answer to. If I’m aware of something, I wouldn’t reach out to my twin. I imagine myself tired and unable to think properly, then I think about how I would ask my twin and how it should help me in that moment.

Generic advice doesn’t help me

When I want to understand something, I want an example I can picture and act on.

For example, our team was working on the Tsuin banner for an upcoming expo. I wrote “we are not a memory tool” in the header. Then I learned that people ignore the negative word, like when someone says there’s no pink elephant in the room and everyone immediately pictures a pink elephant. So people would just read “memory tool”.

Direct answer first

What taught me the most is reading my own answers. When I explained my reasoning to a question, it felt confusing, and it didn’t directly answer the question.

If one day I asked my twin a question and it went round and round before giving me the direct answer of what I can do, I would be pissed off. That shows me I might be unconsciously explaining things in a hard way for someone to understand in my daily life.

Now I’m aware of it, and I actually have space to improve how I explain things to someone.

What would make me say the training works

I’ll say the training works when my twin replies with my reasoning and sounds like me. I think the best sign is when it brings back a lesson I once forgot, and that augments my thinking.