The same half-sentence, five times.

To see for yourself that a model does not store an answer. It builds one, word by word, and the path can differ every time.

Any AI chat tool you already use. Somewhere to write five lines: paper, or the table below.

1. Open a new chat. Not a continued one: an earlier answer would influence the next.
2. Paste exactly this, and nothing else:
   Complete this sentence in one line: "The most surprising thing about working in enterprise technology is"
3. Copy the reply into row 1 of the table.
4. Close the chat. Repeat steps 1–3 four more times, filling rows 2 to 5.
5. Optional: do the whole thing again in a different tool and compare.

Five different endings, some close, some not. Nothing was looked up. Each time the model chose a likely next word, then the next, and the path diverged. If two runs came out identical, that is worth noting too; Note 03 explains why.

write down what you noticed. that's the whole practical.