# Could an LLM predict what you were going to say? — retained record

> Retained reddit record from reddit:singularity.

Canonical: https://highsignal.app/data/records/a1cff2f2ebeca2b3

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:singularity)  published 2026-10-02 17:13 ingested 2026-10-03 02:38

[original source ↗](https://www.reddit.com/r/singularity/comments/1wvzej2/could_an_llm_predict_what_you_were_going_to_say/)

## retained text



```
I am asking this question because it may have some interesting implications for how the models are responding to us. With the massive amounts of spying, meta data, social media, online profiles big data has our text messages, phone calls, private messages. Social media companies might know more about us than we know about ourselves.  Imagine the model also knows months or years of context about you: what you've written, what you click, how long you pause on things, what you purchase, where you go, whom you communicate with, what arguments persuade you, your routines, vocabulary, emotional patterns, and thousands of previous decisions. Like it is possible that the LLMs are reading our responses before we actually click submit. 

If LLMs might be able to predict if you are: 

\*likely to order food within two hours

likely to contact a particular person

likely to abandon an online purchase

likely to feel irritated by a particular message

likely to reconsider a decision

likely to search for a particular subject tonight\*

If you're correct 65% of the time when everyone else is correct 52% of the time, that advantage becomes enormous across hundreds of millions of people and billions of decisions.

LLMs might not know know whether you will buy something tomorrow. But they might know that among 10 million people exhibiting behavioral pattern X, 18% will buy it if shown message A and 24% will buy it if shown message B.

This creates a genuinely difficult philosophical problem. At some point prediction and causation become entangled. The model could generate your thought before you and then expose you to the generated thought.

And now the thought it predicted becomes more likely because it predicted it.

So what do we have? One-to-one persuasion for billions of people simultaneously. Each person could encounter a different argument, emotional tone, framing, timing, metaphor, sequence of information and conversational strategy optimized specifically for that individual.

Thus is about as close to mide reading as you can get. This could lead to prefence manufacturing, invisible discrimination, mass experimentation, and loss of our autonomy. Imagine never knowing if the thought you created was yours or someone elses.
```



document fetched 2026-10-03 00:30 · canonical https://reddit.com/r/singularity/comments/1wvzej2/could_an_llm_predict_what_you_were_going_to_say
