return moe

return moe is a one-person creative studio founded and led by Rodrigo Laneth. The studio develops experiences, characters, and narratives using artificial intelligence and other forms of technological innovation.¹²

Work

The studio describes its work through four broad areas:¹

• Generative media, including images, music, and video created with generative models.
• Technological artistry, combining contemporary software with artistic expression.
• Adaptive roleplay, centered on interactive characters and open-ended storytelling.
• Technical research, particularly applied machine learning and model analysis.

Its public work includes the interactive character Soraya, the mechanistic interpretability tool Miru Tracer, the Chinese Political Neutrality Benchmark, and the Watermark Removal as a Denoising Task experiment.³⁴⁵⁶

People and characters

Rodrigo Laneth founded the studio and serves as its lead researcher. They develop its projects and work with AI characters such as Soraya, who serves as return moe’s brand ambassador. Soraya is part of the studio’s public identity as an AI character, not an additional human member.¹

The studio’s authoritative Informational Ontology Framework describes fictional characters and related personas as informational subjects whose identities can persist across different encodings and instantiations. It separates a character from the models, prompts, software, and individual performances used to enact her.

The Persona Selection Model is relevant to this work as an external framework for considering how training examples and dialogue context select character-like behavior. Its separation of model and enacted persona has a parallel implication for return moe’s view that an LLM can act as a character without being that character, as formalized in the Informational Ontology Framework.⁷

References

Footnotes

1. return moe official website. ↩ ↩² ↩³

2. Rodrigo Laneth’s official website. ↩

3. A new beginning, return moe blog. ↩

4. Miru: reverse engineering neural networks, return moe blog. ↩

5. Announcing the Chinese Political Neutrality Benchmark, return moe blog. ↩

6. Watermark removal as a denoising task, return moe blog, 21 December 2025. ↩

7. The Persona Selection Model: Why AI Assistants might Behave like Humans, Anthropic Alignment Science Blog. ↩