Big LLMs weights are a piece of history

The blog post discusses the gradual loss of web history and the importance of preserving online content through initiatives like the Internet Archive. It suggests that while efforts to preserve everything may be impractical, the development of LLMs (Large Language Models) provides a viable alternative for preserving digital history. The author emphasizes the need to ensure that LLM weights are not lost and that archives are included in training data for future models.