The TechCrunch AI glossary

Artificial intelligence (AI) is a vast and intricate field. Researchers in this domain often use specialized jargon to describe their work. As a result, we frequently incorporate these technical terms when covering developments in the AI industry. To help clarify these concepts, we have created a glossary defining key terms and phrases that commonly appear in our articles.

We will update this glossary regularly to include new entries as researchers continue to push the boundaries of AI while addressing emerging safety concerns.

AI Agent

An AI agent is a tool that utilizes AI technologies to perform complex tasks on behalf of a user. These tasks go beyond basic chatbot functions and may include filing expenses, booking reservations, or even writing and maintaining code. The concept of an AI agent implies an autonomous system capable of executing multi-step tasks by integrating multiple AI systems. However, as the field evolves, interpretations of what constitutes an AI agent may vary, and the infrastructure to support their full capabilities is still under development.

Chain of Thought

Humans can answer simple questions instinctively, but more complex problems often require step-by-step reasoning. For example, solving a puzzle about the number of chickens and cows on a farm might involve writing an equation to find the answer.

In AI, chain-of-thought reasoning involves breaking down a problem into intermediate steps to improve the accuracy of the final result. This approach typically takes longer but produces more reliable answers, especially for logic-based or coding tasks. Chain-of-thought reasoning is enhanced in specialized models through reinforcement learning.

(See: Large Language Model)

Deep Learning

Deep learning is a subset of machine learning where AI algorithms are designed using artificial neural networks (ANNs) with multiple layers. This structure enables the model to identify complex patterns in data without requiring human engineers to define these features manually. Through repetition and adjustment, deep learning systems improve their outputs over time.

Deep learning models require large datasets (millions of data points or more) and significant computational resources. Although they are more capable than simpler machine learning algorithms, they are also more expensive and time-consuming to train.

(See: Neural Network)

Fine-Tuning

Fine-tuning refers to the process of further training an AI model to optimize its performance for specific tasks. This is typically achieved by providing the model with new, specialized data relevant to the desired area of focus.

Many AI companies use fine-tuning to adapt large language models (LLMs) for particular industries or applications, enhancing their utility with domain-specific knowledge.

(See: Large Language Model)

Large Language Model (LLM)

A large language model (LLM) is a type of AI model that powers AI assistants like ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, and Mistral’s Le Chat. These models interpret user inputs and generate responses by predicting the most likely next word in a sequence based on extensive training on text from books, articles, and other sources.

LLMs are built on deep neural networks with billions of parameters (or weights) that capture relationships between words and phrases. This allows them to generate coherent and contextually appropriate responses across a wide range of topics.

(See: Neural Network)

Neural Network

A neural network is the foundational algorithmic structure behind deep learning and generative AI technologies like large language models. Inspired by the interconnected neurons in the human brain, these networks process data through multiple layers, enabling complex pattern recognition.

Although neural networks were conceptualized in the 1940s, advancements in graphical processing units (GPUs) have recently enabled the training of more sophisticated models. This has led to breakthroughs in areas such as voice recognition, autonomous navigation, and drug discovery.

(See: Large Language Model)

Weights

Weights are the numerical parameters that determine the importance of various features in the data used to train an AI model. These values influence how the model interprets inputs and generates outputs.

During training, an AI model starts with randomly assigned weights. As it processes more data, the model adjusts these weights to improve its accuracy. For instance, in a model predicting house prices, weights might be assigned to features like the number of bedrooms, the presence of a garage, or whether the property is detached, reflecting their impact on property value.