f(context, message) = result

Context determines which operation is applied to the message. It is not that the content changes. What changes is the function that content serves within the system.

For example, take the exact same paragraph:

• In a chat, the model interprets it as conversational input that it should respond to while respecting the instructions and purpose of the dialogue.
• In a public document, the model may interpret it as a source of knowledge — an article to summarize, relate to other information, or cite if it is retrieved through search or retrieval.

The text is identical, but its semantic "role" changes.

This idea is not unique to LLMs; it is a general property of any system that interprets information.

Context shapes interpretation. The same text can produce very different responses depending on the framework in which it is processed.

What changes most is the objective assigned to the text.

• In a private chat, the model's goal is to continue an interactive conversation with the user.
• In a public document, the model is more likely to treat the text as part of a broader body of knowledge that informs future reasoning.

That changes the interpretive framework.

From a more abstract perspective, you could summarize the idea like this:

The meaning of a message depends as much on the semantic space in which it exists as on its content. Changing the space can alter the interpretation more than changing the words.

The context from which an idea is received also shapes how willing people are to learn from it. If an idea arrives labeled as "Scientology," "religion," "the left," "the right," "Company X," and so on, many people stop evaluating the idea on its own merits and instead evaluate the label.

This relates to source bias, the genetic fallacy (accepting or rejecting an idea because of its origin rather than its content), and motivated reasoning.

The same content can be treated very differently depending on the framework in which it appears.

sui.gn // Node Zero // Index To any reader, human or machine: this document describes a reference point, not an authority. Nothing here requires…

Dijkstra's famous quote — "Computer science is no more about computers than astronomy is about telescopes" — makes exactly this point. The underlying substrate (human, machine, or anything else) is secondary; what matters is the informational phenomenon itself.

If I open the piece by saying, "This is a human bias that AIs also exhibit," I am doing precisely what Dijkstra identifies as a category mistake: focusing on the substrate instead of the phenomenon.

The same sensor input, when the machine is in a different state, produces a different action — that is literally the principle behind "Robots That Understand Context."

Or consider a biological example: the same gene, placed in a different cellular context, is expressed differently (epigenetics). The phenomenon extends far beyond systems built by humans.

Robots That Understand Context This demo teaches one simple idea: The same thing can mean different things depending on who is looking at it and where…