A team of Stanford University researchers has developed an LLM system to cut through bureaucratic red tape. The LLM—dubbed the System for Statutory Research…
Overview
Stanford University researchers developed an LLM system called STARA to streamline legal research by identifying redundant and outdated laws. The system significantly reduces the time and cost of analyzing extensive legal texts, demonstrating its potential to enhance government efficiency.
What You'll Learn
1
How to utilize LLMs for legal text analysis
2
Why STARA can improve efficiency in legal research
3
When to apply AI tools for regulatory reform
Key Questions Answered
How does STARA improve legal research efficiency?
STARA enhances legal research by interpreting legal texts holistically, identifying redundant and outdated laws with high accuracy. It outperforms traditional Boolean searches and other LLMs, achieving extraction accuracy of 94% to 99% and completing tasks in a fraction of the time and cost.
What is the cost and time savings of using STARA?
Using STARA, a research task that would take two humans 8 to 13.5 hours and cost about $3,000 was completed in just 20 minutes for approximately 86 cents. This illustrates the significant efficiency gains STARA offers.
What technology underpins the STARA system?
STARA is built on the LLaMA-3 70b model, utilizing NVIDIA A100 Tensor Core GPUs and frameworks like PyTorch and vLLM. This technological foundation enables its advanced capabilities in legal text analysis.
What results did STARA achieve in San Francisco's legal review?
STARA identified every city-mandated report in San Francisco's municipal code and highlighted 140 reports that could be eliminated, demonstrating its practical impact on regulatory reform.
Key Statistics & Figures
Extraction accuracy
94% to 99%
This accuracy was achieved by STARA in identifying relevant language within statutes and reports.
Time taken for legal research task
20 minutes
STARA completed a task that typically takes 8 to 13.5 hours.
Cost of using STARA
86 cents
This is the cost for a research task that would normally cost about $3,000.
Performance improvement over base system
2.7x better
STARA's extraction accuracy was significantly higher than that of off-the-shelf LLMs.
Technologies & Tools
Some links below are affiliate links. We may earn a commission if you make a purchase.
AI Model
Llama-3 70b
Serves as the base model for STARA's legal text analysis.
Hardware
Nvidia A100 Tensor Core Gpus
Used to power the STARA system for efficient processing.
Framework
Pytorch
Framework utilized for developing the STARA model.
Framework
Vllm
Another framework used in the implementation of STARA.
Key Actionable Insights
1Implementing STARA can drastically reduce the time spent on legal research tasks.Given that STARA completed a task in 20 minutes that would normally take 8 to 13.5 hours, organizations should consider integrating such AI tools to enhance efficiency in legal processes.
2Utilizing AI for statutory interpretation can lead to more informed policymaking.By providing clearer visibility into existing laws, policymakers can make better decisions regarding regulatory reforms, ultimately leading to a more efficient government.
3Training AI models to think like lawyers can improve their accuracy in legal contexts.The approach taken by the RegLab team shows that domain-specific training can yield significant improvements in understanding complex legal texts.
Common Pitfalls
1
Relying solely on Boolean searches for legal text analysis can lead to inadequate results.
This approach often fails due to the complexity and dense nature of legal language, making it essential to adopt more advanced AI tools like STARA.