The Florida Bar Guide to Getting Started with AI
The Florida Bar offers this guide as a resource for lawyers exploring the use of AI in their law practice. Because this guide is intended as a starting point, it does not attempt to provide a comprehensive set of answers. Each lawyer should explore and make the decision whether to use AI or not based on their individual practices and circumstances, being mindful of applicable ethical rules as well as any unique risks from using particular AI models.
What is Artificial Intelligence (AI)?
Artificial Intelligence (AI) is the simulation of human intelligence in machines programmed to think, learn, and problem-solve like humans. AI performs tasks such as understanding language, recognizing patterns, and making decisions. An AI program is also called a "model." Common AI terms include:
Generative AI: A type of AI model that creates new content, such as text, images, or music, based on large amounts of training data. It uses neural networks to predict the next word or element based on context. Examples for text include ChatGPT, Microsoft’s Copilot, Google’s Gemini, Meta’s Llama, and Anthropic’s Claude, and for images ChatGPT, Gemini, Adobe and Midjourney. A generative AI model that is text-based is called a large language model (LLM).
Hallucination: The generation of incorrect or nonsensical information by an AI model, occurring when the AI produces outputs not based on input data or context, leading to factual inaccuracies.
Multimodal AI: AI that can process and understand multiple types of data simultaneously, such as text, images, audio, and video.
Neural Network: A computing system inspired by the human brain's network of neurons, consisting of interconnected layers of nodes (neurons) that process data, enabling the AI to recognize patterns and learn from data.
Prompt: An input or instruction given to an AI model. The more detailed and specific the prompt, the better the AI’s response or output.
How does generative AI Work?
Generative AI uses neural networks trained on massive amounts of data to recognize patterns. When given a question or prompt, the model uses these learned patterns to predict the best outputs, such as text or images. Training a generative AI model takes months because of the enormous amounts of data and the neural network’s process of learning and optimization. Because AI predictions are just highly customized guesses, they can be wrong.
Researchers are working on the next level of AI, which will allow it to reason and use logic.
What is the difference between AI models?
There are general and law-specific AI models. General models are trained on large sets of human-created data, while legal models take a general model and fine-tune it using law-specific data, such as court opinions, law review articles and example documents. Legal models usually have constraints on the sources of information they use in creating their responses, which are intended to reduce hallucination risk.
Example General AI Models:
- OpenAI’s ChatGPT: Assists with drafting emails, brainstorming, and generating content.
- Google’s Gemini: Can be used to create first drafts and improve existing documents.
- Microsoft’s Copilot: Integrates with Microsoft applications to provide real-time assistance, including coding and writing.
- Anthropic’s Claude: Creates more conversational type writing, summarizes, edits, and translates to other languages
Law-Specific AI Models:
- CoCounsel by Westlaw: Assists with legal research, case preparation, document review, and document drafting.
- Lexis+AI: Assists with legal research, memo drafting, and document review.
- Vincent AI by vLex: Assists with legal research, case preparation, document review, and document drafting.
Where to get started?
Start by exploring general AI models such as ChatGPT, Gemini, or Copilot for administrative tasks and brainstorming. Then, try a free trial of a legal AI model such as CoCounsel, Lexis+AI, or Vincent AI for legal research and document review. Understanding a model's capabilities and limitations through hands-on experience is essential. For now, do not use any confidential or client-specific information.
Run your first prompt
Below is a link to Google’s Gemini (no sign-in required) and an example prompt. Use copy/paste or retype the prompt in the chat box and submit it.
Prompt: I’m a Florida lawyer and want to write a LinkedIn post about Usury under Florida law. Please write the post for me.
Congratulations on running your first prompt! Did you notice how the tone and format of the Gemini result was designed for LinkedIn, including hashtags at the bottom and a call-to-action?
Now that you see how prompting works, try another prompt, this time for a practice area you know well. Importantly, try varying your prompts, such as requesting an answer “less than 250 words” or “written at a 10th grade level.” Revising a prompt with new information or requests changes the output. Remember to not give any client confidential information when first using an AI model.
What about ethics?
Florida was the first state to issue an ethics opinion regarding the use of AI. Florida Bar Ethics Opinion 24-1 provides a good overview of the ethical issues involved in using AI and should be reviewed before using client information in your prompts.
To ensure client confidentiality, you should confirm that you maintain ownership of any uploaded data, learn how to delete your data, and review the AI provider’s license terms or representations regarding confidentiality. General-purpose AI models may use your questions and uploaded documents to train future models unless you have a paid subscription that allows you control over your data.
When you find a general AI vendor you like, check its security reputation, hallucination risk, various AI model features, and paid plan options for individuals or businesses. One advantage of the three main Legal AI models is their promise to maintain client confidentiality, use of strong security, and reduced hallucination risk. However, always verify AI-generated outputs yourself to ensure accuracy and reliability, as AI should assist, not replace, human judgment. And if associates or nonlawyers will be using AI in your firm, consider a user training program and written guidelines for proper AI usage for client matters.
Examples to get started
To begin using AI, identify your specific needs, such as drafting documents, conducting legal research, managing client intake, or creating presentations. Start small, gradually incorporating AI into your practice, and continually verify AI-generated outputs for accuracy and reliability. As AI models continually improve, new specialized AI tools will emerge tailored to various practice areas.
General AI Tasks:
- Drafting administrative letters or marketing articles: Use ChatGPT, Copilot, or Gemini to create initial drafts quickly.
- Generating summaries, creating formulas, customizing presentations: Use Microsoft Copilot to generate summaries of non-legal documents or articles, create formulas in Excel, and custom backgrounds and text in PowerPoint. The same can be done using Gemini in Google Workspace.
Legal AI Tasks:
- Legal research: CoCounsel from Westlaw, Lexis+AI, and Vincent AI from vLex, streamline research by offering relevant case law, statutes, and sometimes secondary references.
- Document review: Use CoCounsel and Lexis+AI for reviewing contracts and legal documents, document comparisons, identifying key clauses, and suggesting improvements.
- Document drafting: CoCounsel Drafting and Lexis Create will draft client-specific documents using their respective form libraries and drafting notes.
- Case preparation: Certain AI models can help organize case files, highlight important information, and suggest strategies.
By thoughtfully incorporating AI into your practice, you can enhance productivity while maintaining high ethical standards and protecting client confidentiality.
Comparison Charts
Comparison Chart of General vs. Legal AI Models*
| General AI models | CoCounsel from Westlaw | Lexis+AI | Vincent AI by vLex | |
|---|---|---|---|---|
| Emails/letters | Yes | Limited | Limited | Limited |
| Marketing (ideas, copy) | Yes | No | No | No |
| Presentations | Yes (e.g., PowerPoint with Copilot, Google Slides) | No | No | No |
| Legal research | Varies by AI model and has hallucination risks. | Yes** | Yes** | Yes** |
| Research memos | Varies by AI model and has hallucination risks. | Yes** | Yes** | Yes** |
| Single document review | Yes and has hallucination risks. | Yes** | Yes** | Yes** |
| Large document or database review | Varies by AI model | Yes** | Yes** | Yes** |
| Document drafting | Yes | Yes, using WL forms | Soon (Lexis Create) | Soon |
| Confidentiality | Only certain paid licenses promise prompts and uploads will not be used for training or stored. | Yes (Learn more) | Yes | Yes |
| Live support help | Limited | Yes | No | No |
*Disclaimer: This chart is based solely on the representations of each provider. AI is continuously evolving so this information is subject to change. The Florida Bar makes no promises or representations about any features and does it endorse any particular AI model.
**All Florida attorneys have an ethical duty of competence to review AI output. Hallucination, misstatement and omission can occur with any AI model. See Florida Ethics Opinion 24-1.
Comparison Chart of Largest Generative AI Models*
| Chat GPT o1 | Microsoft Copilot | Claude 3.5 | Gemini 1.5 | |
|---|---|---|---|---|
| Access to Internet | Yes | Yes | No | Yes |
| Creates images | Yes | Yes | No | Yes |
| Runs code | Yes | No | No | Yes |
| Does data analysis | Yes | Yes, in Excel only | No | Yes |
| Sees images | Yes | Yes | Yes | Yes |
| Sees video | No | No | No | Yes |
| Reads files | Yes | Yes | Some | Yes |
| Integrated with word processing apps | No | Yes | No | Yes |
| Personality | Eager to help, bit bland, verbose. Vocal mode has more personality. | Friendly, can end conversation early. | Strong personality, willing to play along. | It wants to help. A lot. |
| Best for… | The best overall, general model. Vocal mode is impressive and easy to use. | Works directly with Office programs and Win11 in interesting, useful ways. | Excellent writer, high levels of “insight”. | Works directly with Google Workspace apps. Largest context window by far. |
*Chart compiled from information published on "One Useful Thing" Substack. All statements are from that author and subject to change as model abilities evolve.
Additional Resources
- Doing Stuff with AI: Opinionated Midyear Edition is a useful and entertaining review, as of June 2024, about current and upcoming features of the large generative AI models and some special models (i.e., video, music).
- Josh Kubicki’s Generative AI Blueprint course. This online course for lawyers explains how generative AI works, prompt techniques, and use cases. (Approved by The Florida Bar for 4.0 CLE hours.)
- Jeff Su’s Google’s AI Course for Beginners is a short, easy to follow summary of a longer Google online course (YouTube).
- ChatGPT plan comparison chart. Confidentiality is enabled by default on Teams and Enterprise plans. On the Pro plan, the default is to share data, so confidentiality is not enabled unless you opt-out. You can opt-out of data sharing on a paid Pro plan.
- Perplexity is an AI alternative to Google searches by answering the search question (with cites to source links).
- ABA AI Ethics Opinion.
- ABA Task Force on Law and Artificial Intelligence - Year 1 Report (August 2024).