[The Art of Delegating to AI] You can only get back what is visible | AI's Field of View

Have you ever felt that even when you ask an AI the same thing, the quality of the response varies significantly depending on the situation?

One moment it gives you a spot-on answer, and the next, it returns something incredibly vague, even though you haven't changed how you ask.

That is likely not due to the AI's mood, but rather a difference in what it could 'see' at that moment.

So, what does an AI see when it writes a response? Today, I will lay those things out.

👀 Responses depend more on what is visible than on 'ability'

An AI constructs its response using what it can see at that exact moment.

Therefore, even with the same AI, if the materials you show it are different, the response you get back will change. Sometimes, 'what you are showing it right now' has more of an impact than 'whether this AI is smart'.

If you think it's a problem of intelligence, you'll want to change the AI itself. But if it's a problem of what is visible, you can change it yourself. That is where there is a bit of hope.

🗂 The 5 things AI generally sees

When you list them, they generally fall into these categories.

• The instructions and premises you just provided. What you communicated at that time.

• What it remembers. Things discussed previously or preferences (this was the topic of the last article).

• Materials on hand. Text you pasted or files you provided.

• The conversation so far. The flow of the chat you currently have open up to this point.

• Available tools. Searching, opening files, etc. What an AI can do varies.

The combination of these five things is the AI's 'field of view.' The response is created from within this.

🔦 What is not visible cannot be included

It may seem obvious, but if it cannot see something, it cannot include it in the response. Even if it is in your head, if you haven't provided it, it is the same as non-existent to the AI.

The opposite is also true. If it sees unnecessary things, it gets pulled in that direction, and the conversation goes off track.

So, choosing what to show and what not to show. That is, in short, what I am trying to do in this series. It is not about difficult settings, but about what you place in its field of view.

✍️ Just one thing for today

In the middle of a conversation with the AI you are using, try asking this. Whether it is ChatGPT or Claude.

It will probably list a few things. The instructions provided, the conversation so far, the documents pasted, and what it remembers. That is its current field of view.

What I want you to look at is whether that is less or more than you expected. Whether something you wanted included is missing. Once you understand that, you will also see what you need to add next.

It won't necessarily answer everything honestly. If you can find that interesting, including that fact, then that is enough for today.

🍵 In closing

Next time, the topic is "Context is finite (quantity and cost)." Now, if you are wondering if showing it more is always better, it seems that is not necessarily the case. We will explore that.

You can choose what to show yourself. I feel like just knowing that was a harvest for today.

However, deciding what not to show is actually surprisingly more difficult than deciding what to show.

Well, that is fine.

Afterword

Thank you for reading this far.

I touched upon what AI is looking at when it provides an answer.
I believe this is actually an important matter.
It is really troublesome because sometimes what it is looking at changes without you realizing it.

Please feel free to make use of this.

It would encourage me if you could "Follow" or "Like"!
I look forward to your continued support!