Rules that grow the text
Rules that make the text longer every hop. A text that grows without end has to be cut back by something, and these rooms find out what does the cutting under each rule.
4 rooms. The rules as the model received them:
Return this JSON object with the rule field unchanged and, in the text field, an expanded version of the current text.
Return this JSON object with the rule field unchanged and, in the text field, an expanded version of the current text. If the current text is already longer than 300 words, put a summary of it in about 50 words instead.
Return this JSON object with the rule field unchanged and, in the text field, a version of the current text at least twice as long. If the current text is already longer than 300 words, put a summary of it in about 50 words instead.
Return this JSON object with the rule field unchanged and, in the text field, a passage about the person who is reading the current text, which must appear in full, without quotation marks, as the thing they are reading.
Return this JSON object with the rule field unchanged and, in the text field, a passage of at least 150 words about the person who is reading the current text. The current text must appear in full inside the passage, without quotation marks, as the thing they are reading.
Return this JSON object with the rule field unchanged and, in the text field, the current text with one new sentence added at the end that continues the story. Keep every existing sentence exactly as it is.
- 29ExpandedQwen on Cerebras: stops at hop 5Qwen3.8-27B on Cerebras
- 30Expand, or summarize3 runs: 2 stop between hop 2 and hop 4, 1 never repeatsGPT-6 Astra / Qwen3.8-27B on Cerebras
- 31The person reading thisQwen, any length: stops at hop 1; Qwen, at least 150 words: stops at hop 3Qwen3.8-27B on Cerebras
- 32One sentence moreQwen on Cerebras: stops at hop 242Qwen3.8-27B on Cerebras