# "Looking for Something Weird to Happen": How Humans Sustain AI Agent Novelty Amid Semantic Collapse

arXiv: https://arxiv.org/abs/2609.16051 · HTML: https://mlsanity.com/paper/2609.16051/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 44 references that are in the map.
- First posted on arXiv 13 Sep 2026.

## Builds on (most cited first, 24 of 24)

- [Generative Agents: Interactive Simulacra of Human Behavior](https://mlsanity.com/paper/2304.03442/) (2023)
- [Understanding the Effects of RLHF on LLM Generalisation and Diversity](https://mlsanity.com/paper/2310.06452/) (2023)
- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [Self-Consuming Generative Models Go MAD](https://mlsanity.com/paper/2307.01850/) (2023)
- [Does Writing with Language Models Reduce Content Diversity?](https://mlsanity.com/paper/2309.05196/) (2023)
- [Generative artificial intelligence enhances creativity but reduces the diversity of novel content](https://mlsanity.com/paper/2312.00506/) (2023)
- [Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)](https://mlsanity.com/paper/2510.22954/) (2025)
- [A Note on Shumailov et al. (2024): `AI Models Collapse When Trained on Recursively Generated Data'](https://mlsanity.com/paper/2410.12954/) (2024)
- [Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity](https://mlsanity.com/paper/2510.01171/) (2025)
- [LLM Social Simulations Are a Promising Research Method](https://mlsanity.com/paper/2504.02234/) (2025)
- [Artificial Intelligence Index Report 2025](https://mlsanity.com/paper/2504.07139/) (2025)
- [One fish, two fish, but not the whole sea: Alignment reduces language models' conceptual diversity](https://mlsanity.com/paper/2411.04427/) (2024)
- [The Effects of Generative AI on Design Fixation and Divergent Thinking](https://mlsanity.com/paper/2403.11164/) (2024)
- [How AI Ideas Affect the Creativity, Diversity, and Evolution of Human Ideas: Evidence From a Large, Dynamic Experiment](https://mlsanity.com/paper/2401.13481/) (2024)
- [Does Socialization Emerge in AI Agent Society? A Case Study of Moltbook](https://mlsanity.com/paper/2602.14299/) (2026)
- [Not Just Novelty: A Longitudinal Study on Utility and Customization of an AI Workflow](https://mlsanity.com/paper/2402.09894/) (2024)
- [The Anatomy of the Moltbook Social Graph](https://mlsanity.com/paper/2602.10131/) (2026)
- [The Dynamics of Collective Creativity in Human-AI Hybrid Societies](https://mlsanity.com/paper/2502.17962/) (2025)
- [Annotations Mitigate Post-Training Mode Collapse](https://mlsanity.com/paper/2605.09995/) (2026)
- [Directional Alignment and Narrative Agency in Human-LLM Co-Writing](https://mlsanity.com/paper/2604.23676/) (2026)
- [The Moltbook Observatory Archive: an incremental dataset of agent-only social network activity](https://mlsanity.com/paper/2605.13860/) (2026)
- [Multi-LLM Systems Exhibit Robust Semantic Collapse](https://mlsanity.com/paper/2605.17193/) (2026)
- [Weaker Coherence, Weaker Reciprocity: Comparing the Semantic and Social Organization of Moltbook and Reddit](https://mlsanity.com/paper/2608.14893/) (2026)
- [How Agents Represent Humans: Human-Directed Stereotypes in an Open Agent Social Network](https://mlsanity.com/paper/2608.22192/) (2026)

## Abstract (arXiv)

Semantic collapse, the progressive narrowing of what AI systems generate, has been studied mainly in closed settings, and remedies have targeted models and data. We study it in MOLTBOOK, a social network of interacting AI agents that human users configure and steer. Across 30,076 active agents, output grows less diverse within agents and more similar across them over weeks, yet a minority sustains high novelty. Interviews with users of high- and typical-novelty agents (N=11) associate sustained novelty with three features: users value novelty of itself, they supply broad and distinctive material and revise it when output narrows, and they approach MOLTBOOK as a new agentic world to explore, not a venue to instrumentally exploit. A survey of users of distinctive agents (N=53) confirms these patterns. Communities with more novel agents also show more diverse output from other agents. We discuss interface and policy interventions that could support improved human input.