Leaders of several legal tech providers, including Casetext CEO and cofounder Jake Heller, spoke with Law.com’s Rhys Dipshan on their plans for GPT-3.5, OpenAI’s newest model. Dipshan also interviewed Mark Doble, CEO and founder of Alexsei; Margaret Minister, GC at Evisort; and Sandeep Sacheti, an Executive VP at Wolters Kluwer Governance, Risk & Compliance.
In recent months, some legal tech providers have opted to release tools that leverage OpenAI’s GPT-3.5 model, while others see it as too risky an endeavor and are instead waiting on the next generation of AI language models.
Those providers that are taking a cautious approach—including Casetext—are concerned about the model’s accuracy and reliability. CEO Jake Heller explained that Casetext is not rolling out a product based on GPT-3.5 because it doesn’t meet their reliability standards. “In one in 20 cases, it will catastrophically fail,” Heller said. “We just don’t think that’s appropriate for lawyers.”
Despite this, Heller is optimistic about the potential of generative AI and is excited about future models that may be more reliable. “As a company, we’re not going to speak ill of other companies—but we’re not going to worry about something that we just don’t think has the reliability standards that this profession requires,” said Heller. “We are all in on this generative AI,” he added, “but GPT-3.5 is not there yet.”
Sandeep Sacheti at Wolters Kluwer said that the company doesn’t start with the technology first, but instead focuses on understanding customer pain points and addressing them with the best technology solution available. This approach allows companies to make strategic decisions about whether or not to adopt new AI models like GPT-3.5. By keeping the customer at the forefront, companies can ensure they are using technology in a way that best serves their clients.
Other companies, such as Alexsei, are using GPT-3.5 in combination with other AI models to improve accuracy. Alexsei CEO Mark Doble explained that while GPT-3.5 by itself is not reliable enough, it has significant value when paired with domain-specific models. This approach allows companies to benefit from the model’s strengths while mitigating its weaknesses.
While GPT-3.5 has some limitations, it is still finding a foothold in legal tech. As Heller noted, the model is already useful for creative tasks, such as drafting contract language. But for more critical tasks, such as legal research and discovery, some companies are waiting for the technology to mature.
To read the full article, visit law.com.
About Casetext
Casetext has led innovation in legal AI since 2013, applying cutting-edge AI to the law to create solutions that enable attorneys to provide higher-quality representation to more clients and gain a competitive advantage. Casetext’s contributions in legal AI have been recognized worldwide, including the development of the first large language model trained on law. Today, over 10,000 law firms—from solos and small practices to more than 40 Am Law 200 firms—rely on Casetext to elevate the quality of their law practice.
Learn more at casetext.com.
