Developer uses AI digital twin to tune self

This is a pioneer example of using LLMs for "AI-mediated self-actualization," leveraging the technology not just for productivity but for personal character and style alignment.

• –**A "Digital Twin" for Style:** By providing a model with a massive 38K token dataset of their best writing, the author creates a simulation of their "idealized self" to use as a benchmark.
• –**Anti-Sycophancy Strategy:** Utilizing models like Big-Tiger-Gemma-27B-v3 (known for being an anti-sycophancy fine-tune) ensures the AI provides harsh, objective criticism of their real-world actions rather than simple agreement.
• –**Human-in-the-Loop RLHF:** The author is effectively performing "Reinforcement Learning from Human Feedback" on themselves, with the AI acting as the reward model that highlights the gap between their actual and ideal behavior.
• –**Local LLM Utility:** Demonstrates a high-value use case for local models (Gemma-2, GLM-4) that allows for processing sensitive personal writing history without privacy concerns.