Reproduce “LAPIS: Language Model-Augmented Police Investigation System” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Heedou Kim, Dain Kim, Jiwoo Lee, Chanwoong Yoon, Donghee Choi, Mogan Gim, Jaewoo Kang
LAPIS assesses Korean criminal-investigation hypotheses using a locally fine-tuned 10.8B language model and retrieved legal premises. For each True/False hypothesis, it adds five relevant premises and returns an assessment with rationale. On 1,229 Korean police exam-derived test hypotheses, LAPIS reaches 0.74 total accuracy, versus GPT-4’s 0.73; without retrieval, its accuracy is 0.66. These results concern exam judgments, not field performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.