The realm of artificial intelligence is complex and multifaceted. To navigate this field, researchers and experts often employ specialized terminology and jargon. In order to provide clarity and understanding, we have compiled a comprehensive glossary of key terms and phrases commonly used in our coverage of the artificial intelligence industry. This glossary will be regularly updated to incorporate new entries, reflecting the ongoing advancements and discoveries in the field of artificial intelligence, as well as emerging safety concerns.

As researchers continue to push the boundaries of artificial intelligence, identifying novel methods and potential risks, we will expand this glossary to include new terms and definitions, ensuring that our readers remain informed and up-to-date on the latest developments in the field.

An AI agent is a sophisticated tool that utilizes artificial intelligence technologies to perform a range of tasks on behalf of the user, exceeding the capabilities of a basic AI chatbot. These tasks may include filing expenses, booking tickets or reservations, or even generating and maintaining code. However, the concept of an AI agent is still evolving, and different individuals may have varying interpretations of its meaning. As we have previously explained, the development of AI agents is a dynamic and rapidly changing field, with ongoing infrastructure development and refinement of their capabilities. At its core, an AI agent represents an autonomous system that can draw upon multiple AI systems to execute complex, multi-step tasks.

When presented with a straightforward question, the human brain can often provide an answer without requiring extensive thought or calculation. For instance, determining which animal is taller, a giraffe or a cat, is a relatively simple task. However, more complex problems may necessitate the use of pen and paper to arrive at a solution, as they involve intermediary steps. For example, if a farmer has a combination of chickens and cows, totaling 40 heads and 120 legs, solving this problem may require writing down a simple equation to determine the number of each type of animal (20 chickens and 20 cows).

In the context of artificial intelligence, chain-of-thought reasoning for large language models involves breaking down a problem into smaller, intermediate steps to enhance the accuracy and quality of the final result. Although this process may take longer to produce an answer, it is more likely to yield a correct solution, particularly in logic or coding contexts. Reasoning models, which are developed from traditional large language models, are optimized for chain-of-thought thinking through the application of reinforcement learning.

(See: Large language model)Deep learning is a subset of self-improving machine learning, characterized by the use of artificial neural networks (ANNs) with multiple layers. This structure enables deep learning algorithms to identify more complex correlations and patterns in data compared to simpler machine learning-based systems, such as linear models or decision trees. The design of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain, allowing them to learn and improve over time.

Deep learning AIs are capable of identifying important characteristics in data without requiring human engineers to define these features explicitly. The algorithms can learn from errors and refine their outputs through repetition and adjustment. However, deep learning systems require large amounts of data to produce accurate results, often involving millions of data points. Additionally, training deep learning algorithms typically takes longer and is more expensive than training simpler machine learning models.

(See: Neural network)

Fine-tuning refers to the process of further training an AI model to optimize its performance for a specific task or area, typically by feeding it new, specialized data. This process involves building upon the existing training and knowledge of the model, allowing it to adapt and improve its capabilities for a particular application or domain.

Many AI startups are leveraging large language models as a foundation for developing commercial products, seeking to enhance their utility and value for a target sector or task. To achieve this, they supplement the initial training of the model with fine-tuning based on their own domain-specific knowledge and expertise, enabling the model to better address the needs and challenges of that particular area.

(See: Large language model (LLM))

Large language models, or LLMs, are the AI models used by popular AI assistants, such as ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, or Mistral’s Le Chat. When interacting with an AI assistant, users engage with a large language model that processes their requests directly or with the assistance of various tools, such as web browsing or code interpreters.

It is worth noting that AI assistants and LLMs can have distinct names. For instance, GPT refers to OpenAI’s large language model, while ChatGPT is the AI assistant product.

LLMs are deep neural networks consisting of billions of numerical parameters (or weights) that learn the relationships between words and phrases, creating a representation of language as a multidimensional map of words. These models are trained on vast amounts of data, including billions of books, articles, and transcripts, which enables them to identify patterns and generate text.

When prompted, an LLM generates the most likely pattern that fits the input, evaluating the most probable next word based on the context. This process is repeated to produce a coherent and contextually relevant response.

(See: Neural network)

A neural network refers to the multi-layered algorithmic structure that underlies deep learning and the broader realm of generative AI tools, which have emerged in the wake of large language models. The concept of neural networks, inspired by the human brain’s interconnected pathways, dates back to the 1940s. However, it was the relatively recent development of graphical processing hardware (GPUs), driven by the video game industry, that unlocked! the true potential of neural networks. These specialized chips have enabled the training of algorithms with many more layers, leading to significant performance improvements in various domains, including voice recognition, autonomous navigation, and drug discovery.

The term “weights” is central to AI training, as it refers to the numerical parameters that determine the importance or relevance of different features or input variables in the training data. Weights shape the AI model’s output by applying multiplication to the inputs. The training process typically begins with randomly assigned weights, which are then adjusted as the model seeks to produce an output that closely matches the target.

To illustrate this concept, consider an AI model designed to predict house prices based on historical real estate data. The model may include weights for various features, such as the number of bedrooms and bathrooms, the type of property (detached, semi-detached), and the presence or absence of parking or a garage. The weights assigned to each of these features reflect their relative influence on the property’s value, as determined by the training data.

Ultimately, the weights that the model attaches to each input feature provide insight into the relationships between these factors and the property’s value, based on the specific data set used for training.