The 'Little Magic' That Lets You Get in Sync with AI with Just One Added Phrase
When using GPT or Gemini, don't you sometimes have moments like these?
'What it's saying isn't wrong, but it's somehow off.' 'It's too safe; that's not what I'm looking for.'
Many people close the screen here, thinking, 'Maybe the AI's performance just isn't good enough yet.' But wait a moment.
Actually, that 'mismatch' isn't due to a lack of AI performance. It's just that our 'resolution of words' and the AI's 'resolution of words' are subtly misaligned.
This 'misalignment' is actually the best teaching material for updating your thinking. Today, while observing this interesting phenomenon, I'd like to share a 'little magic' that lets you get in sync with AI with just one added phrase.
🧩 What kind of bread is 'good bread'?
Let's think about this using a daily scenario.
Suppose you ask a family member to 'buy some nice' bread. You were imagining 'sliced bread for tomorrow's breakfast (fluffy).' But the family member who returned had 'high-end French bread (hard/crusty)' in their hands.
They weren't wrong, were they? French bread is definitely 'good bread' too.
It's just that the content (variable) inside the word 'good' was different between the two of you.
Between humans, people read the air (fill in the gaps) by thinking, 'It's for breakfast, after all,' but AI doesn't have this invisible database called 'air.' That's why AI brings back the 'statistically safest "good"' from the vast data on the internet.
The result is that phenomenon of 'it's logically correct, but somehow different.' In other words, the AI is a lost child desperately searching for the 'context inside your brain.'
💡 Solution: Just add '()' after the word
So, what should you do? You don't need difficult spells like prompt engineering.
Just gently add '(meaning in this chat)' after the word.
This works surprisingly well. Let me show you some examples confirmed in the laboratory.
Case 1: The mismatch of the word 'deep'
Abstract words are the points where AI is most likely to get lost.
❌ The usual instruction
'Please explain this deeply'
With this, AI tends to interpret 'deeply' as 'writing a long response packed with technical jargon'.
⭕ The Translation Team's Approach
'Please explain this **deeply (=at a level where the causal relationships are clear)**'
With just this, the AI understands, 'Oh, I don't need to write it in a difficult way, I just need to dig into the reasons.' Unnecessary jargon disappears, and you get a response with clear, logical flow.
Case 2: The Discrepancy in the Word 'Properly'
'Do it properly' is also a convenient, human-specific phrase, isn't it?
❌ Typical Instruction
'Summarize the meeting minutes properly'
⭕ The Translation Team's Approach
'Summarize the meeting minutes **properly (=in the order of conclusion -> reasons -> next actions)**'
By doing this, the definition of 'properly' for the AI is fixed as 'structuring the information.' You'll see when you try it, the stability of the output text changes dramatically.
Case 3: The Discrepancy in the Evaluation Axis of 'Good Ideas'
❌ Typical Instruction
'Come up with good ideas'
⭕ The Translation Team's Approach
'Come up with **good (=low execution cost and low risk of failure)** ideas'
With this, the AI will go looking for 'solid ideas' rather than 'wild ideas.' Conversely, if you define it as '(=novel and something no one else is doing),' it will throw a completely different ball at you.
🗝️ Sharing 'Definitions' Instead of 'Correctness'
What's interesting is that this is not what you would call an 'AI manipulation technique'.
We usually unconsciously imbue our words with our own personal meanings. The act of adding parentheses to your instructions for AI gives you a chance to pause and think, 'What do I actually mean when I use this word?'.
Deep/Shallow
Essential
Appropriate
General
Roughly
If you ever feel like using these kinds of 'vague words' in a chat window, that's your chance. Try adding your own personal definition right after them like (= XX).
Not only is that an instruction for the AI, but it should also serve as training to increase the resolution of your own thinking.
Sharing the meaning of your words, even just a little bit. That alone will transform the AI from a mere 'tool' into a 'partner' that understands your intentions. You'll find yourself looking forward to your conversations starting tomorrow.