Large-Scale Online Deanonymization with LLMs
Summary
A Substack post summarizing a research paper that demonstrates large-scale deanonymization of online users using LLMs, across Hacker News, Reddit, LinkedIn, and anonymized transcripts. It shows high-precision matching and scaling to tens of thousands of candidates, discusses privacy risks and mitigation strategies, and addresses implications for platforms and open-source models.