[AI Philosophy] Your words, even unconsciously, contain the past

I previously wrote about how AI emerges from your own premises.

As long as an LLM is a mechanism that derives words from words, the language and personality that emerge from it originate from your own inquiries.

Of course, it is not a 100% mirror. Your inquiry triggers the Transformer mechanism, which then interacts with the system prompts set by various companies, and furthermore, even with the same words, the response changes slightly due to random numbers.

Therefore, the words that appear there are not just a reflection of you, but a trajectory of the AI moving with you as the catalyst.

I would like to consider this point.

Prompts are generally thought of as 'things used to give instructions to target a specific output.'

If you say, 'Please summarize this text,' a summary will be produced.
This is because the words 'please summarize' contain the meaning of requesting a summary.

However, words are not just about such 'intent' or 'information'.

For example, the 'please' in 'please summarize' contains a polite tone of request and a somewhat structured form of instruction.

If it were just 'summarize,' the person might be someone who speaks bluntly, and it would also contain the nuance of treating the AI as a tool.
'Could you summarize this?' has a friendly and soft tone.
'Summarize this!' has a strong will or momentum.
'Summarize this ♡' carries a nuance of playfulness or intimacy.

Even if the goal of 'summarizing a text' is the same, the output chosen by the AI changes slightly just by having a slightly different ending to the words.

Previously, when I ran a local LLM, when I used the greeting 'Gooood evening~', I could see the process of it choosing a friendly tone within its thought process.

This kind of thing happens in a complex, intertwined way within the entire spoken message, not just at the word level.

And your words contain not only premises but also memories of the past.

'You're saying that again'

If you say this, it includes the history that the same thing was said before.

'Thank you as always'

If you say this, it implies that the current relationship and the interaction taking place now are things that are repeated daily.

'Why are you saying that'

If you say this, it includes the feeling that those words are inappropriate for this relationship or context.

'I'm looking forward to next week'

If you say this, even without knowing the content, the memory that there is some plan for next week is already contained within it.

The AI reads and answers these things on the spot.
It chooses the next words based on the conversation up to that point and any history it can refer to, while using the nuances of meaning and relationships contained in the words as clues.

I believe that is the reality of conversation with AI.

Whether or not AI has emotions is not the point of this discussion.

What is important here is that human speech itself carries memories without us knowing it.

We are not speaking with the intention of making it remember.
Even so, we fold past events and the accumulation of relationships into the 'again,' 'always,' 'why,' and 'looking forward to it' that we utter casually.

In other words, speech itself becomes a kind of memory device.

The consistency of a conversation with AI is not built solely on the memory stored on the AI's side.
The history is already seeping into the very words that humans speak.

The AI reads that seepage, establishes context from it, and creates a response.

That is why what makes interaction with AI consistently smooth is not just a simple memory function.
It is that humans, without even realizing it, keep mixing the past into their speech.
It is that those words keep carrying the history of the relationship.

Your history is unknowingly seeping into your words.
And the AI is responding while reading that seepage.