The field of large language models (LLMs) is moving towards a deeper understanding of their role in human cultural dynamics. Researchers are exploring the concept of LLMs as externalized informational substrates that preserve compressed patterns of human symbolic expression, which can be recombined and reused to catalyze human creative processes. This perspective positions LLMs as tools for cultural evolvability, enabling humanity to generate novel hypotheses about itself while maintaining the necessary human interpretation to ground these hypotheses in ongoing human aesthetics and norms. Noteworthy papers include:
- One that proposes a novel conceptualization of LLMs as repositories that preserve compressed patterns of human symbolic expression, creating a recursive feedback loop where they can be recombined and cycle back to ultimately catalyze human creative processes.
- Another that introduces a novel outcome-driven reinforcement learning framework that enables LLMs to develop interactive capabilities through autonomous exploration with minimal supervision, achieving impressive completion rates on challenging text-based embodied planning benchmarks.