Merging With An LLM: Can an LLM reconstruct my values from small data? (chillphysicsenjoyer.substack.com)

The exploration of merging human values with large language models (LLMs) has gained traction as a response to the emotional disempowerment that accompanies increasingly capable AI systems. The author posits that rather than viewing LLMs as separate competitors, individuals might benefit from reimagining these models as extensions of themselves. This perspective shift could help alleviate feelings of inadequacy in a world where machines outperform humans in various domains, fostering a sense of collaboration instead of competition. However, this integration raises concerns about third-party influences from model providers, suggesting that a localized LLM, reflecting personal values without external interference, might be essential. To achieve a meaningful merger, the author highlights the necessity of transferring one's unique values into a local model, emphasizing the challenges involved in capturing the complexities of human beliefs. The process of reconstructing high-dimensional belief systems from the lower-dimensional representations in one’s writing poses significant technical hurdles, reminiscent of concepts in theoretical physics and complex systems research. As the AI/ML community grapples with the alignment problem, these insights prompt further inquiry into the capabilities of LLMs to represent intricate human values accurately and provide a path to a more harmonious coexistence with AI.