https://lqxiao.github.io/blog.html Blog posts by Liqiang Xiao on LLM alignment, safety, self-play, and reinforcement learning. en https://lqxiao.github.io/post.html?p=llm-value-investing https://lqxiao.github.io/post.html?p=llm-value-investing Wed, 12 Nov 2025 00:00:00 GMT It's easy to imagine LLMs outperforming humans in high-frequency trading: machines are faster, tireless, and consistent. Whether LLMs make good value investors, however, remains an open question. https://lqxiao.github.io/post.html?p=llm-safety https://lqxiao.github.io/post.html?p=llm-safety Thu, 17 Jul 2025 00:00:00 GMT After two years working on LLM safety, I’d like to distill what I’ve learned and look toward the next horizon. I hope this post helps anyone tackling safety in an industrial setting. Although academia and industry share … https://lqxiao.github.io/post.html?p=llm-self-play https://lqxiao.github.io/post.html?p=llm-self-play Thu, 29 May 2025 00:00:00 GMT Self-play is a kind of self-evolutional method that a large language model (LLM) generates its own prompts and responses in order to improve its performance through interaction with itself. Instead of relying solely on h… https://lqxiao.github.io/post.html?p=llm-hallucination https://lqxiao.github.io/post.html?p=llm-hallucination Thu, 02 Jan 2025 00:00:00 GMT Hallucination is usually defined as model generates nonsensical or unfaithful content.