Bibliographic record
Abstract
The integration of artificial intelligence (AI) and large language models (LLMs) has significantly influenced numerous industries. However, the legal sector remains cautious, given its stringent demands for confidentiality, accuracy, and adaptability. Over the summer, I conducted research examining the intersection of law and LLMs, focusing on their current limitations and potential advancements. My analysis revealed that even the cutting-edge models (e.g., ChatGPT, Claude) exhibit notable deficiencies in legal reasoning. Specifically, they struggle to: Accurately identify legal issues, Retrieve relevant case law reliably, Apply legal principles to the facts with precision, and Provide accurate pinpoint legal citations. To address these gaps, I investigated how lawyers analyze legal issues and craft responses in the form of legal opinions. This research informed the development of a structured system prompt designed to enhance OpenJustice, Queen’s open-source legal AI. Within OpenJustice, I assisted with the research on the Code for Dialogue (CoDial) framework and its dialogue flow system. My work involved analyzing how lawyers approach complex legal issues, how lawyers deconstruct legal principles into smaller components, and translating these principles into a structured graph for AI interpretation. In collaboration with the engineering team, I provided support for the implementation of functionalities that enable dialogue flows to resemble a legal reasoning process. Additionally, I contributed to research exploring an LLM-as-a-Judge framework, which aims to evaluate AI-generated legal responses and identify areas for improvement. Preliminary findings indicate that evaluating legal responses presents unique challenges, as legal questions often lack a singular correct answer; two valid, reasonable, and well-supported arguments may lead to opposing conclusions. This work highlights the complexities of integrating AI into the legal domain and underscores the importance of continued research. Advancements in this field hold significant potential to enhance access to justice and improve efficiency within the legal profession.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.015 | 0.004 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.004 | 0.006 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".