AI Glossary: Decoding Key Terms from AGI to Hallucinations

Artificial intelligence is a fascinating yet complex field, often accompanied by specialized vocabulary that can seem opaque to those unfamiliar with the subject. To make this domain more accessible, a glossary of common AI terms is regularly updated to include the latest advancements and challenges, particularly in terms of security.

General Artificial Intelligence (AGI)

General artificial intelligence, or AGI, is a concept that sparks numerous interpretations and debates. OpenAI's CEO, Sam Altman, described AGI as the "equivalent of a median human you could hire as a colleague." According to OpenAI's charter, AGI refers to highly autonomous systems that outperform humans in most economically valuable work. On the other hand, Google DeepMind views AGI as AI that is at least as capable as humans in most cognitive tasks. This diversity of interpretations reflects the complexity and ambiguity surrounding the concept of AGI.

AI Agent

An AI agent is a tool that uses artificial intelligence to accomplish a series of complex tasks, such as managing expenses, booking tickets or tables at a restaurant, and even writing and maintaining code. This concept goes beyond the capabilities of traditional chatbots. However, the term "AI agent" can have different meanings depending on the context, as the infrastructure necessary to fully realize its capabilities is still under development. The central idea is that of an autonomous system that can rely on multiple AI systems to execute multi-step tasks.

Chain of Thought

In the context of artificial intelligence, chain of thought reasoning refers to the ability of models to break down a problem into intermediate steps to improve the quality of the final outcome. For example, to answer a complex question, an AI model may be prompted to follow a series of logical steps to arrive at a correct answer. This process is particularly useful in logical or coding contexts. Reasoning models are developed from traditional language models and optimized for chain thinking through reinforcement learning, which generally takes longer but increases the likelihood of obtaining a correct answer.

Computation

The term "computation" in the context of AI refers to the processing power essential for AI models to function. This computational power is provided by hardware such as GPUs, CPUs, TPUs, and other forms of infrastructure that form the backbone of the modern AI industry. This type of processing is crucial for training and deploying powerful models, and it is often used as a shorthand to refer to the types of hardware that provide this power.

Deep Learning

Deep learning is a subset of machine learning that uses a structure of artificial neural networks (ANN) with multiple layers. This approach allows for more complex correlations compared to simpler machine learning systems, such as linear models or decision trees. Deep learning models are capable of identifying important features in data on their own, without requiring human engineers to define these features. However, these systems require many data points to produce good results (millions or more) and generally take longer to train, leading to higher development costs.

Diffusion

Diffusion is a technology at the heart of many AI models generating art, music, and text. Inspired by physics, diffusion systems "slowly destroy" the structure of data—such as photos or songs—by adding noise until nothing remains. AI diffusion systems aim to learn a "reverse diffusion" process to restore the destroyed data, thereby acquiring the ability to recover data from noise.

Distillation

Distillation is a technique used to extract knowledge from a large AI model via a "teacher-student" model. Developers send queries to a teacher model and record the outputs. These outputs are then used to train the student model, which is trained to approximate the teacher's behavior. This method allows for the efficient transfer of knowledge from a complex model to a simpler model.

Fine-Tuning

Fine-tuning an AI model refers to the additional training of a model to optimize its performance for a task or domain that is more specific than what was previously the goal of its training. This is typically done by feeding new specialized data focused on a task. Many AI startups take large language models as a starting point to build a commercial product but seek to enhance utility for a specific sector or task by complementing previous training cycles with fine-tuning based on their own domain-specific expertise.

Generative Adversarial Network (GAN)

A GAN, or generative adversarial network, is a type of machine learning framework that underpins significant developments in generative AI to produce realistic data—including (but not limited to) deepfake tools. GANs use a pair of neural networks, one generating output from its training data, while the other model evaluates that output. This competitive structure allows for the optimization of AI outputs to be more realistic without additional human intervention.

Hallucination

The term hallucination is used in the AI industry to refer to the phenomenon where AI models generate incorrect information. This poses a major problem for AI quality, as hallucinations can produce misleading results and lead to real-world risks. The issue of AIs fabricating information is often attributed to gaps in training data. Hallucinations contribute to a trend toward increasingly specialized or vertical AI models, meaning domain-specific AIs that require narrower expertise.

Inference

Inference is the process of executing an AI model. This involves allowing a model to make predictions or draw conclusions from previously seen data. It is important to note that inference cannot occur without training; a model must learn patterns in a dataset before it can effectively extrapolate from that training data.

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 Llama, Microsoft's Copilot, or Mistral's Le Chat. These models are the foundation of advanced interactions with AI, enabling a variety of powerful applications in natural language processing and other fields.