8 Adoption

In a rapidly evolving landscape, selecting generative artificial intelligence (AI) systems for adoption is critical. The choice extends far beyond comparing performance benchmarks; it requires careful consideration of factors, including capability, cost structures, terms of service, and specific use case demands.

As these models become increasingly sophisticated and diverse, users and organisations face the complex task of navigating options ranging from powerful proprietary solutions, such as ChatGPT and Claude, to open-source alternatives like Llama and Mistral, or bespoke options. This section aims to introduce you to the leading proprietary platforms that are available. For the well-known proprietary solutions, the various models and their capabilities are outlined.

Some models listed below have been evaluated for their performance in legal use cases.^[1] The LegalBench project is an open science initiative to assess legal reasoning in English LLMs. LegalBench serves as a benchmark comprising various legal reasoning tasks. The models have been tested on functions including:^[2]

• Issue spotting (identifying legal issues in a set of facts)
• Rule recall (identification of relevant rules)
• Rule application (identification of relevant facts and how they influence the outcome)
• Rule conclusion (determining outcome)
• Interpretation (parsing and understanding legal text)
• Rhetorical analysis (reason about legal argumentation and analysis)

The Models