This AI glossary is an excerpt from AI for the Rest of Us, a comprehensive book by Nextpoint and Tom O’Connor that offers honest, practical guidance on AI for legal professionals. Click here to learn more and buy your copy.
The technical terminology surrounding AI is vast and complex. Whether you’re just getting started or need a quick refresher, it’s a good idea to familiarize yourself with some of these concepts before diving deeper into the world of AI. Take a look through this glossary and feel free to return to it as these terms crop up throughout your AI journey.
TERMS INCLUDED
- Artificial Intelligence (AI)
- AI Agent
- Chatbot
- ChatGPT
- Fine-Tuning
- Generative AI (GenAI)
- Hallucination
- Large Language Model (LLM)
- Machine Learning (ML)
- Natural Language Processing (NLP)
- Parameter
- Predictive Coding
- Prompt
- Recall and Precision
- Retrieval-Augmented Generation (RAG)
- Small Language Model (SLM)
- Technology Assisted Review (TAR)
- Token
Artificial Intelligence (AI)
AI is a branch of computer science focused on creating systems that can perform tasks typically requiring human intelligence, such as reasoning, learning, problem-solving, perception, and understanding language. AI is everywhere, from the facial recognition you use to open your smartphone to the algorithmic recommendations you get on Netflix. While current conversations around AI have focused on generative AI systems, the term encompasses a much broader range of technologies that have been in use for decades.
AI Agent
AI agents are systems that can act autonomously to achieve goals; they are designed to make independent decisions and take action without constant human guidance. Unlike traditional AI that responds to specific inputs, AI agents can plan sequences of actions, adapt to changing circumstances, and pursue objectives over time. This technology, known as agentic AI, is still in its early stages. Waymo’s self-driving cars are one of the most advanced real-world deployments of agentic AI today, but this level of automation is not widespread in most industries. Still, AI agents have begun to pop up in legal technology, and this will likely increase as the technology continues to evolve.
Chatbot
A chatbot is a software application designed to simulate human-like conversation with users. Crude chatbots have existed since the 1960s, and they’ve played a key (if often frustrating) role in online customer support for years. The release of ChatGPT by OpenAI in 2022 was a steep change in the capabilities of chatbots. Its combination of a large language model with training that emphasized successful conversations made it much more broadly applicable than any previous chatbot. New versions of ChatGPT, along with competitors such as DeepSeek, Microsoft’s Copilot, Google’s Gemini, and Anthropic’s Claude, have been so successful that to many people, chatbots are simply synonymous with AI.
ChatGPT
ChatGPT is a conversational AI model developed by OpenAI, based on the GPT (Generative Pre-trained Transformer) architecture. It is trained to generate human-like text responses in a conversational manner, useful for a variety of applications like answering questions, providing explanations, and engaging in dialogue. However, legal professionals should exercise caution when using open models like ChatGPT. You have little control over what happens to your data after it is processed by the chatbot, meaning sensitive information should not be shared. Additionally, ChatGPT can generate false or misleading responses that require careful verification.
Fine-Tuning
Most large language models (LLMs) are trained on masses of text that span vast ranges of topics – often whatever the trainer can get their hands on. A few are trained on text that is limited to a particular domain. However, a more common and generally less computationally demanding approach to tailoring LLMs to a particular subject is fine-tuning. This approach starts with a broad domain LLM and uses topic-specific or task-specific data to adjust some of its parameters. In the law, this might mean fine-tuning a generic LLM with documents from a particular law firm, or from a particular task such as contract clause extraction, risk analysis, or compliance monitoring.
Generative AI (GenAI)
As discussed in the foreword, generative AI refers to AI tools or components that create new content, such as text, images, music, or videos, based on learned patterns from existing data. After the launch of ChatGPT in 2022, generative AI became the center of the current AI revolution. GenAI holds huge potential for attorneys, since so much of legal work involves drafting contracts, briefs, and other types of documents.
Hallucination
Hallucinations refer to situations where a generative AI system produces outputs that are factually incorrect, inaccurate, logically incoherent, or completely fabricated. The most notorious examples are when large language models (LLMs) output “facts” that sound plausible but are completely made up. Attorneys using GenAI to draft briefs have earned the ire of judges for fake case citations hallucinated by LLMs. Users of GenAI thus have a responsibility to verify information they get from AI tools, as these hallucinations can lead to dire mistakes if left unchecked.
Hallucinations in the Courtroom
Surely lawyers everywhere learned a lesson from the first high-profile case where an attorney faced sanctions for citing fake cases hallucinated by ChatGPT. Right? Unfortunately, too many lawyers have ignored that lesson and echoed this sloppy use of AI. Legal researcher Damien Charlotin has compiled a database to track legal decisions related to AI hallucinations in the courts. This serves as a crucial reminder to always verify AI outputs before filing materials in a court of law. Explore the database at damiencharlotin.com/hallucinations.
Large Language Model (LLM)
A large language model (LLM) is a type of machine learning model, specifically a generative model, trained on massive data sets of text to replicate patterns in language. They can generate fluent language on almost any topic in response to natural language prompts, making them a key component of chatbots, document generation tools, and many other AI systems. The “large” in LLM refers to the number of parameters in the model as well as the amount of training data used to produce them.
Machine Learning (ML)
Machine learning was traditionally a subfield of AI, but has arguably swallowed the whole field in recent years. ML algorithms allow computers to learn patterns from data and make decisions or predictions without being explicitly programmed. Modern computing systems have access to a vast amount of data about the world, and ML systems have the ability to adapt to changes in the world as reflected in that data. This has led to ML playing the dominant role in modern AI systems.
Natural Language Processing (NLP)
Natural language processing (NLP) is a branch of AI focused on the interaction between computers and human (natural) languages. It enables machines to read, understand, interpret, and generate human language, making it a key component of generative AI models. Research and development in NLP focuses on automating practical tasks involving language, such as document generation, question answering, text and multimedia search, language translation, and many others.
Parameter
Parameters refer to the internal variables or weights in an AI model that are learned during training. Think of parameters as the “knobs and dials” that an AI model tweaks while learning. These parameters help the model make predictions or generate outputs based on the input data it receives. Essentially, they are the settings that the AI adjusts during the learning process to minimize errors and improve its ability to perform tasks. The number of parameters in a model is often used as a measure of the model’s complexity. For example, large models like GPT-3 and GPT-4 have billions or trillions of parameters, which enable them to perform complex tasks such as natural language generation.
Predictive Coding
Predictive coding is a process used in ediscovery where a machine learning algorithm is trained to identify relevant documents by learning from a set of human-reviewed examples. The algorithm then predicts the relevance of the remaining documents. Predictive coding can improve the efficiency and speed of document review and ensure that attorneys focus on the most important material.
Prompt
A prompt in AI is the input text or instruction given to an AI model to guide its response or behavior. It’s essentially how you communicate with the AI system, telling it what you want it to do, what context to consider, or what kind of output you’re looking for. The quality and specificity of the prompt directly influence the relevance and usefulness of the AI’s response. Prompt engineering refers to the task of developing and refining prompts in order to generate the desired outputs.
Recall and Precision
In ediscovery and legal AI tools, recall and precision are two key metrics for evaluating how well a system can identify relevant documents. Recall represents the percentage of relevant documents retrieved from the body of data, while precision measures the number of documents labeled as relevant that were actually relevant. Put simply, recall answers the question, “Did you find everything?” while precision asks, “How much extra junk did you pick up?” Balancing these two metrics helps ensure that no important document is overlooked while minimizing false positives.
Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation (RAG) refers to generative AI techniques that leverage an LLM alongside a body of task-specific text. In response to a user prompt, a RAG system retrieves relevant information from the specified source to generate a response from the LLM. RAG approaches enable AI to work with specialized information and help reduce (but do not eliminate) hallucinations. RAG forms the foundation of Nextpoint’s AI-powered transcript summary feature.
Small Language Model (SLM)
Small language models are designed for language tasks but trained on a smaller data set and optimized for specific applications, often with fewer parameters than large language models. While not as powerful as LLMs, small language models are still useful for specific legal applications. Because they are smaller, they may offer faster processing and can be more easily customized for niche legal tasks. However, it is important to note that there is no official technical distinction between small and large language models. Different software developers may have varying definitions for how many parameters constitute a “small” vs. “large” model.
Technology Assisted Review (TAR)
Technology Assisted Review (TAR), also known as Computer Assisted Review (CAR), refers to machine learning and AI techniques used to streamline the document review process in ediscovery. These technologies allow lawyers to review large data sets by prioritizing the most relevant documents, thus saving time and reducing human error. The earliest forms of TAR were centered on predictive coding, where attorneys review a small sample of documents and the AI learns from their decisions to classify the rest of the data automatically. In the past two years, TAR systems based on generative AI and prompting have emerged and show promise of producing more effective results with less manual review.
Token
Tokens are the basic units that AI language models use to process text. They can be words, parts of words, or even individual characters, depending on how the text is broken down. For example, the word “tokenization” might be split into tokens like “token” and “ization.” Some AI tools employ token-based pricing, which functions similarly to the per-GB pricing models of traditional ediscovery software.
What’s Most Important for Legal Professionals?
Something as technically complex as AI inevitably brings plenty of complicated jargon to the table. You might feel overwhelmed reading through this list, and that’s okay. You don’t have to memorize each term – this glossary is a resource that you can continuously reference as you navigate the world of AI. For now, here are the themes that are most essential for legal professionals to understand:
Generative AI and LLMs
As we mentioned, GenAI tools have inspired the recent wave of excitement around AI. Naturally, this means that a number of terms in this list tie back to generative AI in some way. You likely don’t need to know the intricacies of natural language processing, but it’s important to understand what GenAI is and how it differs from other types of AI. This can help you wade through the “AI” label that’s slapped on so much new tech we see today. Additionally, you must understand the risks and limitations of GenAI, including hallucinations, to ensure it doesn’t create issues for your practice.
Predictive Coding and TAR
These technologies introduced AI to the legal field in 2010 and are still commonly used to streamline document review. However, these tools require far more manual input than the AI systems employed today. If you haven’t already used TAR for document review, it likely won’t be the first AI tool you jump to use in 2026. But understanding the history of AI in the legal field can help you understand where it’s going as we face a digital revolution.
Fine-Tuning and RAG
These are two methodologies that enable developers to hone AI tools for specific tasks and subject matters like the law. Prompt engineering also plays a role in obtaining particular outputs from an AI system, which can even occur behind the scenes in an AI tool. These approaches can reduce – but not eliminate – hallucinations. AI products that employ these methods are more likely to be beneficial for the specific needs of the legal field.
Parameters and Tokens
Parameters and tokens are two terms that relate to the back-end structure of an AI model. As a legal professional, you’ll rarely need to dive deep into technical concepts like these. However, parameters provide information about the size and complexity of a model, and pricing is often centered around tokens, so it’s important to have some familiarity with these terms in case they crop up in your search for effective AI products.
Bottom line? You don’t need to become a data scientist. But knowing these basics will help you ask the right questions, evaluate new tools, and navigate technology with confidence as AI continues to evolve.
Get Your Copy of “AI for the Rest of Us”
AI for the Rest of Us offers more than this AI glossary — it dives into legal applications of AI, covers the risks and ethics of using AI, and shares practical tips for implementing AI in your legal practice. Click the button below to learn more and buy your copy today.
