GPT-3 in Legal – The Hottest Innovation to Hit Legal Ops

Not even three months old and ChatGPT has disrupted multiple industries across the globe, including legal operations. The craziest part is that ChatGPT is still growing and maturing. In time, as it gets fed more data, more opportunities for leveraging this powerful tool should unlock themselves.

What is ChatGPT

Developed by OpenAI, ChatGPT is an interactive language model that produces human-like text in response to prompts. OpenAI used the GPT-3 language model as the foundation for ChatGPT.

Ever since it launched, the internet has been going crazy over the possibilities of ChatGPT and AI. ChatGPT is able to answer almost any question, and it can assist you with basic tasks like writing an email, essay, or piece of code. It’s even able to conduct research and compile results in an easy-to-read manner.

Popularity of ChatGPT

When it comes to use cases, there are a lot where ChatGPT can excel, especially when it comes to finding extremely specific information. This has made it highly attractive to businesses and individuals as a solution that helps them accelerate and scale their work (admittedly, it’s also fun to see what results AI comes up with).

As an example, imagine you want to find the price per word for a YouTube scriptwriter. Compared to proofreading and editing, YouTube scriptwriting is a relatively new niche industry, which could make finding numbers a bit challenging.

You might be able to spot possible quotes after spending an hour or so digging through multiple articles and job sites. Or you can just query ChatGPT and it will provide ranges (and additional pricing considerations) in a few seconds, saving you the time and mental gymnastics of doing the research yourself.

ChatGPT also excels as a generative AI. When you need to brainstorm titles or topics for an article you’re working on, it can provide several. It’s even able to generate potential talking points about distinct topics, including itself and whether or not it can provide valid legal advice or legal services.

Some have actually experimented with using ChatGPT to code simple games and programs.

In any case, conversations (and uses) of ChatGPT are everywhere now. According to Swiss bank UBS, ChatGPT is the fastest-growing app of all time, taking just two months to reach 100 million users. By comparison, TikTok took nine months to hit the same number of users.

GPT-3 in a Nutshell If You Don’t Follow the News

All of this wouldn’t have been possible without the underlying tech – GPT-3. Compared to ChatGPT, GPT-3 is a much larger, more sophisticated language model. In addition to carrying out a wider variety of natural language tasks, it can perform zero-shot transfer learning (which is a setup where a model can learn knowledge and classify objects without having undergone specific training).

With this in mind, GPT-3 is the more powerful model of the two. But due to ChatGPT being more conversational, user-friendly, and intuitive, ChatGPT AI has become more ubiquitous. This can be partly attributed to the fact that additional data was used to train and adjust GPT-3 for use as a chat application (which would ultimately become ChatGPT).

How is GPT-4 different from GPT-3?

OpenAI has already announced they’re going to release GPT-4 in March. In addition to containing nearly 30 gigabytes more of training data, GPT-4 is multimodal, whereas GPT-3 is unimodal.

Multimodal means that GPT-4 is able to accept both text and images as inputs for generating an output. By comparison, unimodal models like GPT-3 and ChatGPT are limited strictly to text and they don’t accept images as an input.

In addition to the larger amount of training data and multimodal nature, another key difference is the number of parameters that make up the model.

Some estimates indicate that GPT-4 may contain up to 1 trillion parameters. Although OpenAI hasn’t released the exact numbers, the general consensus is that the number of parameters in GPT-4 is at least more than GPT-3’s 175 billion.

The combination of more training data, more parameters, and the flexibility of being multimodal is expected to make GPT-4 produce significantly more accurate results than its predecessors.

But since GPT-4 hasn’t been launched yet, we’ll continue to explore the use of GPT-3 in legal operations.

Training GPT-3 in Legal

ChatGPT is just one outcome of adjusting GPT-3.

GPT-3 was trained using a huge amount of text from various sources such as books, articles, websites, and social media posts. Consequently, it can generate text in a wide array of styles and genres.

However, it can easily be modified for a specific purpose, as OpenAI did with ChatGPT. All it takes is enough examples of a certain type of text to retarget GPT-3 for one’s own purposes.

There are two primary approaches to retraining GPT-3: fine-tuning and feeding.

Training GPT-3 in law: fine-tuning vs feeding

To fine-tune GPT-3, several hundred samples of possible prompts and expected outputs are needed. The more samples there are, the more reliable GPT-3 will be at providing the expected output.

In contrast, feeding is more of a quick-and-dirty method. It’s possible to repurpose GPT-3 with just a few samples of work, but the reliability and predictability of results are significantly lower.

When feeding GPT-3, you first provide the model with a couple of samples. The model will process the samples, look for patterns, and then complete your request based on the information you provided.

It’s faster and not as resource-intensive, but as can be expected, the smaller body of evidence will lead to mixed results. In real-world terms, the return of fine-tuning versus feeding is like a Michelin-star chef versus a home cook.

ChatGPT in Legal Ops

ChatGPT unlocks several possible opportunities for legal operations. Strategic, meaningful implementation of the model could enable legal ops to speed up and scale their contracting processes.

Here are a few major use cases. This list, however, is just the tip of the iceberg:

Clause creation

One drawback of ChatGPT and GPT-3 is that they are limited in terms of the amount of text they can generate. Also, the longer the prompt, the less reliable the result will be.

Consequently, these models aren’t well-suited to drafting an entire contract. However, they are perfect for drafting individual clauses in accordance with specific instructions.

This allows legal teams to quickly create error-free clauses in moments. All lawyers will need to do is indicate the requirements and check the end result.  This can speed up your contract management workflow.

Legalese simplification

For better or worse, legal jargon is dense and difficult to work through. This, in turn, causes bottlenecks and snags in workflows. It also results in a greater work burden for legal as non-legal teams constantly turn to lawyers for explanations.

ChatGPT can simplify legalese so that non-lawyers are able to understand the contents of any legal document. This both speeds up workflows while freeing legal ops from constant questions.

Document search

Another minor limitation is that ChatGPT contains data only up to 2021. This means that as far as ChatGPT is concerned, everything that happened in 2022 “never happened”.

This isn’t a dealbreaker. ChatGPT can be successfully leveraged even if it doesn’t contain all data. It just requires a bit of quality control to guarantee that facts and data are correct.

Still, ChatGPT and GPT-3 can both be used to speed up the search of legal documents. Whether it’s locating simple names or conducting a full-text search, the model can be deployed to cut down on time spent on tedious searches. This enables legal to spend more time on analysis and less on discovery.

If you’re not convinced by ChatGPT’s ability to browse documents, then consider that Microsoft announced that it’s planning to use ChatGPT to power its Teams and Bing platforms.

GPT-3 Features in ContractWorks

ContractWorks is the first contract lifecycle management solution to offer a fully functional contracting solution that allows legal teams to capitalize on GPT-3’s capabilities. Not a prototype, not a beta, but rather a complete product.

ContractWorks currently enables two contract management legal tech features that are built using GPT-3: Clause Creator and Simplify.

Clause Creator

Clause Creator can be found during the contract redlining stage. Simply add any requirements that the clause should include and indicate the target length. After generating the clause, you can review the language and make any edits before inserting it into the contract.

Here’s a sample of what Clause Creator can produce:

Prompt: Generate a non-compete for 90 days for both parties after the contract ends.

Output: Both parties agree not to compete with each other for a period of 90 days after the contract ends.

Clause Creator empowers legal and non-legal units to handle contracts faster and draft clauses more efficiently as clauses can be created and negotiated in seconds, instead of days.

Simplify

The second feature that uses GPT-3 is Simplify, which does exactly as the name suggests – it simplifies language.

To use Simplify, select the text you want to “translate” into simple words. GPT-3 will process the text and produce an easier version that you can copy and send to anyone.

With Simplify, complex legalese transforms into simple English in moments. Here’s a sample of Simplify in action:

Legalese: Mutual Obligations. Each party discloses its Confidential Information to the other under this agreement. “Discloser” will mean the party disclosing Confidential Information and “Recipient” will mean the party receiving Confidential Information from Discloser. Each of the parties is a discloser in respect of Confidential Information owned by such party.

Simple English: Each party will disclose its confidential information to the other party. The party that discloses the information is the discloser and the party that receives the information is the recipient.

Simplify provides non-legal teams the freedom to focus on the meaning and contents of contracts while liberating legal ops from always fielding every single question. With Simplify, legal can merely point non-legal to Simplify for basic explanations. This also streamlines collaborative contract management, enabling efficient cooperation between legal and non-legal teams throughout the contract lifecycle.

Tips for Leveraging ChatGPT

Although ChatGPT has already made a massive splash on current practices, it’s not infallible. Here are some tactics that will help you benefit from implementing the current iteration of ChatGPT:

• If the answer you receive is irrelevant, reword your prompt so that it’s simpler and more specific.
• If the answer you receive continues to be irrelevant, select data to fine-tune GPT-3 or completely rethink the wording of your prompt.
• Don’t use ChatGPT to draft large amounts of text.
• Use keywords to help the model ensure the answer is relevant.
• Keep in mind that ChatGPT contains data up to 2021, not later.
• You’ll need a human to perform a small amount of quality control to ensure data and outputs are accurate.
  

These tips will help you successfully integrate ChatGPT and GPT-3 to accomplish your contracting needs while mitigating any potential downsides of ChatGPT.

To learn more ways to maximize the benefits of ChatGPT, get our guide “How to Train ChatGPT to Become Your Legal Assistant”.

Conclusion

For savvy, digitally literate lawyers, ChatGPT has the potential to completely reshape the way they go about their daily work. However, in its current form, ChatGPT is better suited as a highly intelligent legal assistant who is capable of accomplishing menial, less interesting tasks.

At the moment, most current implementations of GPT-3 in legal operations revolve around text generation and rephrasing. But these are just the earliest days. Given enough time, more GPT-3 (and eventually GPT-4) legal tech features will be launched, such as AI-supported redlining. At ContractWorks contract management, we’re also working on implementing AI-driven redlining features to our Collaborative Editor.

Book a demo with our team to find out how ContractWorks can help you integrate ChatGPT into your contracting.