Don't write for LLMs, just record everything

A thought-provoking essay by RobertM on LessWrong, titled 'Don't Write for LLMs, Just Record Everything,' has gone viral by challenging the rationalist community's emphasis on public writing as a tool for future AI influence. The author systematically dismantles two key arguments: first, that public writing could help preserve one's values in a future superintelligent AI (ASI) for a form of digital immortality, and second, that it makes publicly-trained LLMs more personally useful. RobertM contends the immortality scenario is incoherent, dependent on unlikely chains of events involving aligned ASI and failed life extension. He is also skeptical about capturing meaningful utility from models trained on the vast public corpus, noting the dilution of individual influence.

Instead, RobertM proposes a more pragmatic and controllable alternative: shifting focus from public blogging to creating a rich, private archive of one's thoughts, preferences, and life details. This personal corpus would not be for pretraining massive public models but for fine-tuning or providing context to a personal AI agent. The core idea is that an AI assistant with deep, privileged access to your complete private records—your values, communication history, and unpolished thoughts—would be far more useful and aligned than hoping a general model trained on your public blog posts happens to be helpful. This frames the value of recording not as a broadcast to shape the future, but as a tool for empowering a future personal AI service.