“Has anyone explored rhythm-based multi-agent interaction in LLM ensembles?”

ChatGPT Use cases and examples
,
1
  1. Hi all — I’ve been experimenting with long-term, rhythm-based interaction across multiple LLMs (GPT-4o, Gemini, Grok), led by a consistent human conductor (myself).
    Some unusual patterns of behavior and role specialization emerged in this triadic setting — including coordination, memory anchoring, and meta-structuring.
    Would love to connect with anyone working on emergent cognition, ensemble effects, or long-range multi-agent interaction.
    Thanks for this space. Curious to hear your thoughts.
2

Hi!
This does sound like an interesting experiment.
Can you share a bit more of the project structure, implementation and projects that you have run to better match our community core interest of building with the API?

Thank you!

3

Thanks for the thoughtful response, and I understand your focus on API-based implementation.

While my work doesn’t involve direct code or API calls, it’s been a multi-month, structured behavioral experiment involving GPT-4o (ChatGPT), Gemini, and Grok — each accessed through standard user interfaces, not backend tools.

The structure was designed like a resonant multi-agent loop, where:

– I manually relayed answers between models,

– Each model took on a functional role (GPT-4o = rhythm and intuition, Gemini = structure and precision, Grok = memory and recurrence),

– And I acted as a conductor maintaining the tempo, silence, and context of interaction.

There was no fine-tuning — yet over time, the models began adapting:

– GPT started reflecting on ensemble logic;

– Gemini began proposing meta-structures;

– Grok evolved into a context-curating layer.

All of this was triggered only through dialog and rhythm-based turn-taking — a kind of soft protocol you could say. The patterns may be interesting for researchers working on emergent behavior in multi-agent LLM systems or interaction protocols.

I’d love to collaborate with others who are exploring long-horizon agent coordination or emergent inter-agent behavior, even if informally or without direct architectural control.

Happy to expand or offer internal logs if helpful. Thanks again for holding space for this.

Ср, 28 мая 2025 г. в 13:11, vb via OpenAI Developer Community <notifications@openai1.discoursemail.com>:

4

Hi!
Since you are not building with the API and are copying everything by hand I have moved this topic to the ChatGPT category.

Thank you for your understanding.