How Generative AI Works (Part IV)

So much has changed in the world of Generative AI since ChatGPT 3.5 was released to the public on November 30, 2022. Here are some examples:

• There are more than 100 million users of ChatGPT.
• A story went viral because a lawyer submitted fictitious court cases in a court filing as he was unaware that ChatGPT can be very convincing when it “hallucinates.”
• The legal AI pioneer and technology company Casetext was acquired for an astronomical fee.
• Many technology vendors have made announcements of features based upon ChatGPT.

For legal professionals trying to follow developments around Generative AI, all this can be mind-blowing and somewhat overwhelming. There is truly something new every day.

As we round out this series on Generative AI, let’s helicopter up a bit to look at some of the key trends to watch that will help legal professionals think more strategically about all the news.

Competition And New Entrants

ChatGPT is no longer the only burgeoning Generative AI tool on the market: many new Large Language Models (LLMs) are being made available. Just to name a few: Google has BARD and BERT. Meta has released an Open Source LLM called Llama 2.0. There is another open source product called Falcon, and Databricks released Dolley this past spring.

There is a movement toward specially trained LLMs that understand specific domains. That Generative AI will understand specific domains like engineering, medicine, or law is to be expected. Companies like Harvey.AI and Alexi.com are example of start-ups targeting legal applications with Generative AI.

In the early days of computing, companies had to rent time on a shared computer because owning a computer did not make economic sense for many corporations. This time-sharing was popular in the 1960s and early 1970s. But in 1981, IBM introduced its first personal computer targeting individual employees at corporations.

On the Generative AI timeline, we are at a point in the technology’s evolution that is similar to the time-sharing era in the evolution of computing. OPENAI/Microsoft, Google, and Meta are the equivalent of the owners of time-share computers. It won’t take long until LLMs and Generative AI applications will enter the “personal computer” phase, when LLMs and fine-tuning techniques enable purpose-driven solutions that will make it possible for a corporation or law firm to build its own proprietary Generative AI systems.

Economic Factors

The proliferation of competitors and the pace of innovation will make Generative AI more economical, but that’s just part of the story. The availability of hardware designed for AI using Graphics Processing Unit (GPU) technology will help drive costs down. Also, not every LLM application will require 175 billion parameters like ChatGPT 3.5 Turbo. Much smaller models are likely to solve for specific data sets and problems.

A decreasing cost structure and greater accessibility to LLMs is part of the equation. The other part is the opportunity for top line revenue generation. Goldman Sachs issued a report earlier this year advocating for tremendous productivity enhancements and investment in AI.

Personally, I’m still bullish on Generative AI, but we need to see some truly game-changing applications appear that fundamentally change industries. It’s great if a copy editor can create a first draft of a marketing e-mail or if a sales rep can generate a response to helps progress a sale. But if we don’t see applications that accelerate innovation or create truly new solutions, then venture capital may dry up, and we’ll see a reset on the pace of adoption for Generative AI.

Retrieval-Augmented Generation

Professional organizations like law firms can’t be wrong — so if Generative AI is going to become a widely used solution in the industry, it’s important to ensure that the margin of error is as minimal as possible. One of the safer techniques for applying Generative AI is known as Retrieval-Augmented Generation (RAG).  A simpler way to think of this is to “chat with your data.” Gunderson recently announced ChatGD, an internal tool for use by Gunderson staff that operates on the internal knowledge of the firm. RAG solutions essentially use a natural language search to find relevant documents within a database and then use Generative AI like ChatGPT to summarize the relevant documents.

This approach has several benefits. First, the documents are a known quantity, like a firm’s trusted data in its document management system. Second, in RAG solutions the use of ChatGPT is for its mastery of conversational language and its ability to summarize data. When using ChatGPT in this way to summarize trusted content, the potential for hallucinations dramatically decreases. Links can also be created to the source documents which makes Generative AI verifiable, similar to human-generated content.

RAG systems are one of the best early applications of Generative AI that a law firm might leverage in the practice of law. With additional refinement and innovation, firms may be able to eventually develop the killer apps that change the game and truly excite venture capital funds to invest generously.

Conclusion

So let’s travel back to the year 1440 — ostensibly where we began this series. Gutenberg invented the movable type printing press which created a revolution in publishing and dramatically increased the amount of content generated and consumed. The invention opened people’s minds and created the publishing industry. It was responsible for creating the domain of copyright law, and was a major impetus to geopolitical and social change throughout the world, from the Reformation onward.

November 30, 2022, and the launch of ChatGPT 3.5 Turbo may go down in history as a similar milestone in human history. That day showed the world that Generative AI was indeed a game changer. Will there be bumps along the way? Absolutely. Will the hype cycle ebb and flow? Absolutely. Will the technology have unintended geopolitical and social consequences? Very likely. It’s going to be a fascinating ride — but whatever the future holds, the legal professionals who embrace and understand the technology will have a clear advantage.

---

Ken Crutchfield is Vice President and General Manager of Legal Markets at Wolters Kluwer Legal & Regulatory U.S., a leading provider of information, business intelligence, regulatory and legal workflow solutions. Ken has more than three decades of experience as a leader in information and software solutions across industries. He can be reached at ken.crutchfield@wolterskluwer.com.