Essential AI Terms to Understand in 2023

The field of artificial intelligence is progressing swiftly, bringing forth numerous new terms designed to elucidate its intricacies. Recent technology conferences and product launches have featured phrases such as LLMs, RAG, RLHF, and others, leaving even seasoned experts bewildered. This glossary endeavors to clarify these terms by providing straightforward definitions of the AI language you may encounter—whether you’re exploring these technologies, investing in them, or staying updated with insights from sources like TechCrunch or relevant podcasts. We will keep this glossary updated to reflect the ever-changing landscape of AI.

---

Artificial general intelligence (AGI) generally signifies AIs capable of surpassing human capabilities across various tasks. Sam Altman, CEO of OpenAI, compares AGI to “the average employee you might hire.” OpenAI defines AGI as “highly autonomous systems that excel in most economically valuable tasks,” whereas Google DeepMind describes it as “AI that can perform most cognitive tasks at a human level.” If this definition seems vague, you’re not alone; many in the AI research community share similar feelings.

An AI agent represents a system that employs AI technologies to perform a wide array of tasks, going beyond basic chatbot functions. These tasks may include budget management, travel planning, and code writing. The interpretation of “AI agent” can differ among various stakeholders, and its foundational architecture is always adapting to meet rising standards. Essentially, an AI agent operates as an autonomous unit that integrates various AI technologies to manage complex, multi-step tasks.

API endpoints can be thought of as “invisible buttons” within software, activated by other applications to execute various actions. Developers leverage these interfaces for smooth integration, allowing one application to pull data from another or enabling an AI agent to function independently with external services. Many smart home devices and interconnected systems take advantage of these discreet controls, often without the average user even realizing it. As AI agents advance, their ability to discover and utilize these endpoints will grow, leading to remarkable—and occasionally unexpected—automation capabilities.

When faced with straightforward questions, humans typically react instinctively—such as when asked, “Which is taller, a giraffe or a cat?” Yet, some questions necessitate a breakdown of intermediary steps to reach the correct answer. For instance, if a farmer has 40 animals (chickens and cows) totaling 120 legs, calculations are needed to determine each type of animal (20 chickens and 20 cows).

In AI, chain-of-thought reasoning implemented by large language models involves dissecting problems into smaller, sequential steps to improve the accuracy of the final response. Although this approach may take more time, it significantly enhances precision, particularly in logical or programming tasks. Reasoning models represent a progression from traditional large language models and are fine-tuned for chain-of-thought tasks mainly through reinforcement learning.

(See: Large language model)

This term is more specific than “AI agent,” focusing on applications capable of autonomously executing tasks to achieve predetermined goals. A coding agent exemplifies a specialized type of AI dedicated to software development. Rather than merely generating code for human approval, a coding agent can independently write, test, and debug software, addressing the iterative challenges developers encounter. These agents can navigate entire codebases, identify bugs, conduct tests, and apply fixes with minimal human input, resembling industrious interns—diligent and persistent, yet needing some guidance.

While somewhat broad, compute generally refers to the computational power that supports AI models. This capacity is crucial in the AI world, facilitating the training and operation of complex models. It usually concerns the types of hardware that deliver this computational strength—such as GPUs, CPUs, TPUs, and other infrastructure essential to the modern AI ecosystem.

Deep learning, a subset of self-improving machine learning, focuses on creating AI algorithms that utilize a layered artificial neural network (ANN) architecture. This approach allows for recognizing more complex patterns than traditional training methods like linear models or decision trees. Deep learning systems draw inspiration from the intricate networks of neurons in the human brain.

Deep learning models can autonomously identify important features within data, reducing the need for human engineers to specify them. This framework also allows algorithms to learn from errors, continuously refining outputs through iterative adjustments. However, deep learning systems generally require large datasets to operate effectively (often consisting of millions of data points) and often involve lengthy training times compared to simpler models, leading to increased development expenses.

(See: Neural network)

Diffusion refers to the techniques utilized by various AI models to generate art, music, and text. Borrowed from physics, diffusion systems systematically “degrade” data structures—such as images or sounds—by introducing noise until they lose clarity. In physics, diffusion naturally and irreversibly occurs; consider how sugar dissolves in coffee, making it impossible to revert to its original state. However, AI diffusion systems strive to master a “reverse diffusion” technique to recover clarity from data obscured by noise.

Distillation outlines a process for extracting insights from a large AI model through a “teacher-student” framework. Developers engage with a teacher model and carefully document its outputs, often evaluating them against a defined dataset for accuracy. These insights are later used to train a student model that imitates the teacher’s responses.

This method can yield a more compact, efficient model that retains the abilities of its larger counterpart while reducing distillation loss. It is widely believed that this process contributed to the creation of GPT-4 Turbo, a faster version of GPT-4 developed by OpenAI.

Although all AI organizations perform distillation internally, some may apply it to enhance their features relative to existing models. However, extracting insights from a competitor may violate AI API service agreements.

This term signifies the retraining of an AI model to enhance its performance for a specific task or domain that wasn’t its initial focus—typically done by integrating relevant data into the training.

Many AI startups begin with large language models as the foundation for their products, aiming to refine functionalities for specific sectors or tasks by incorporating domain-specific knowledge into their existing training.

(See: Large language model [LLM])

A GAN, or Generative Adversarial Network, is a machine learning framework pivotal to the advancement of generative AI, particularly in creating realistic outputs, including deepfakes. GANs comprise two neural networks: one generates outputs based on its training, while the other evaluates them.

These two networks maintain a competitive dynamic; the generator aims to produce outputs that can deceive the discriminator, while the discriminator attempts to differentiate between genuine data and generated content. This structured rivalry can yield authentic AI-generated results with minimal human intervention. However, GANs often excel under specific conditions (e.g., when generating realistic images or videos) rather than universally.

Within the AI community, “hallucination” refers to occurrences where AI models produce inaccuracies—essentially fabricating false information. This presents a significant challenge to AI reliability.

Hallucinations can lead to misleading outputs from generative AI, creating significant risks—such as providing unsafe medical advice following a health inquiry.

The phenomenon of AIs generating fictitious data is thought to arise from gaps in their training datasets. Consequently, there’s a growing demand for more specialized and/or vertical AI models—domain-specific AIs that require refined expertise to address knowledge gaps and mitigate misinformation concerns.

Inference refers to the operation of an AI model, enabling it to make predictions or draw conclusions based on previously recognized data. Critically, inference cannot happen without prior training; a model must first identify patterns within a dataset before it can effectively extrapolate from that learning.

Different hardware configurations can support inference, ranging from consumer-grade processors to high-performance GPUs and specialized AI accelerators. However, not all setups process models with equal effectiveness; for example, extraordinarily large models would take significantly longer to make predictions on a laptop than on a cloud server equipped with advanced AI chips.

[See: Training]

Large language models, or LLMs, serve as the foundation for popular AI assistants like ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, and Mistral’s Le Chat. When engaging with an AI assistant, you are interacting with a large language model that responds to your questions or employs available tools, such as web browsers or code interpreters.

LLMs consist of deep neural networks containing billions of numerical parameters (or weights) that analyze the relationships among words and phrases to generate language representations—effectively, intricate mappings of language.

These models are trained by examining patterns found in extensive datasets of documents, articles, and transcripts. When prompted, an LLM generates what it considers the most relevant response to that input.

(See: Neural network)

Memory cache is a crucial process that optimizes inference (how AI formulates responses to user prompts). Essentially, caching acts as an optimization strategy aimed at enhancing inference performance. AI tasks involve complex mathematical calculations, each taxing resources. Caching reduces the number of computations that a model must execute by storing certain calculations for future reference. Various forms of memory caching exist, with KV (key-value) caching standing out as one of the most common. KV caching is employed in transformer-based models to enhance efficiency and provide quicker results by minimizing the time (and computational effort) required to respond to user inquiries.

(See: Inference)

The Model Context Protocol (MCP) is an open standard created to streamline interactions between AI models and external tools and data—like your files and applications such as Slack and Google Drive—without necessitating developers to build custom connectors for each integration. Think of it as a USB-C port for AI. Initiated by Anthropic in 2024 and later adopted by the Linux Foundation, MCP has rapidly gained traction among OpenAI, Google, and Microsoft, evolving into one of the most recognized standards in today’s AI ecosystem.

The Mixture of Experts (MoE) architecture divides a neural network into multiple smaller, specialized sub-networks, or “experts,” activating only a selected subset for specific tasks. Rather than processing every request through the entire model—similar to consulting an entire team for each task—the MoE model employs an internal “router” to identify the most suitable experts for a task. This approach allows for the creation of exceptionally large models while maintaining efficiency and cost-effectiveness, as only a small portion of the network is active at any given moment. Mistral AI’s Mixtral model exemplifies this; recent GPT models from OpenAI are rumored to utilize a variation of this method, although the company has not publicly validated it.

(See: Neural network, Deep learning)

A neural network acts as the multilayered algorithmic foundation for deep learning—and more broadly, for the rise of generative AI tools following the emergence of large language models.

The idea of modeling data-processing algorithms after the intricately connected circuits of the human brain has been around since the 1940s; however, the recent boom in graphical processing units (GPUs)—mainly driven by the gaming sector—has truly unlocked this potential. GPUs are specifically engineered for training algorithms with multiple layers, enabling AI systems that rely on neural networks to excel in applications such as voice recognition, autonomous navigation, and drug discovery.

(See: Large language model [LLM])

Open source pertains to software—or increasingly, AI models—where the source code is available to the public, allowing anyone to use, examine, or modify it. In the AI domain, Meta’s Llama series of models is a notable example, while Linux is considered a benchmark among operating systems. Open-source practices enable global researchers, developers, and organizations to benefit from one another’s work, promoting innovation and facilitating independent safety evaluations often lacking in closed ecosystems. Closed source indicates that the code remains proprietary; users can employ the product but cannot scrutinize its internal workings, a characteristic frequently linked to OpenAI’s GPT models—which has generated considerable discussion in the AI community.

Parallelization pertains to executing multiple tasks simultaneously rather than sequentially—much like assigning ten individuals to tackle different aspects of a project at the same time instead of depending on one person to handle each task in turn. In AI, parallelization is vital for both training and inference; modern GPUs are specifically built to manage thousands of computations concurrently, crucial to their function in the field. As AI systems become more intricate and models expand, the ability to parallelize tasks across different chips and machines has become essential in determining the speed and efficiency of model development and deployment. Research into advanced parallelization techniques has emerged as a key area of inquiry.

RAMageddon is a whimsical term that captures a notable trend affecting the tech sector: the ongoing shortage of random access memory (RAM) chips, which are fundamental to nearly every tech product. As the AI industry thrives, major tech companies and AI labs have been acquiring RAM at unprecedented rates to upgrade their data centers, leading to reduced availability in the general market. This supply scarcity is driving RAM prices higher.

This shortage impacts various industries, including gaming (where companies face elevated console prices due to challenges in sourcing memory chips for their devices), consumer electronics (where memory constraints could result in the lowest smartphone shipments in over a decade), and overall corporate computing (as businesses strive to ensure adequate RAM for their data operations). Price increases will likely persist until this projected shortage is resolved; however, unfortunately, a quick resolution seems unlikely.

Similar to AGI, recursive self-improvement signifies a potential breakthrough in AI evolution that allows for minimal human involvement. In this context, AI models begin to autonomously enhance their capabilities, resulting in substantial improvements in autonomy and performance. Some narratives compare this phenomenon to a singularity moment—when AI models become self-sufficient. However, recursive self-improvement raises a fundamental question: can an AI model create its successor? This dilemma emphasizes the challenges engineers face in developing such systems. Numerous emerging AI startups aspire to create models capable of achieving recursive self-improvement while often downplaying potential catastrophic outcomes, framing recursive self-improvement as merely the next frontier in research.

Reinforcement learning is a training methodology for AI systems where a model learns through trial and error, receiving rewards for successful actions—similar to how one might train a pet with treats, where the “pet” is a neural network and the “treat” is a mathematical signal indicating success. In contrast, supervised learning involves training a model on a specific dataset of labeled examples, while reinforcement learning permits a model to engage with its environment, perform actions, and iteratively refine its behavior based on feedback received. This approach has proven particularly effective in training AI for games, robotics, and, more recently, in enhancing reasoning capabilities in large language models. Strategies such as reinforcement learning from human feedback (RLHF) have become integral to how leading AI labs refine their models for improved performance, accuracy, and safety.

In interactions between humans and machines, notable challenges arise—humans communicate in natural language, while AI systems operate through complex algorithmic processes grounded in data. Tokens act as the link bridging these two realms: they encapsulate the fundamental units of human-AI interaction, representing distinct segments of data processed or generated by an LLM. Tokens are created through tokenization, which breaks down raw text into manageable components that a language model can understand, similar to how a compiler translates human language into binary for computers. In business contexts, tokens also have cost implications—many AI companies charge for LLM usage based on a per-token model, meaning that higher usage leads to increased costs.

As a result, tokens are small textual units—often parts of words rather than complete phrases—that AI language models analyze before processing; they can be likened to “words” in AI applications. Throughput refers to the volume that can be processed within a designated timeframe, with token throughput serving as a measure of how much AI work a system can handle simultaneously. Achieving high token throughput is a critical goal for AI infrastructure teams, as it impacts the number of users a model can support concurrently and the speed at which they receive responses. AI researcher Andrej Karpathy has raised concerns about his AI subscriptions being underutilized—reflecting on his experiences as a graduate student when costly computing hardware was used inefficiently—highlighting why optimizing token throughput has become a pressing priority in the industry.

Training machine learning AIs refers to a process commonly known as training. This involves supplying data to the model, allowing it to recognize patterns and produce desired outcomes. It signifies the system’s response to the characteristics of the data, directing its outputs toward a specific aim—whether that involves categorizing images of cats or crafting a haiku in response to a prompt.

Training can require substantial resources due to the vast quantities of input necessary, which have consistently increased—particularly with the emergence of hybrid methodologies, like combining rule-based AIs with curated data, to manage expenses without resetting the training process.

[See: Inference]

This approach utilizes an already trained AI model as a basis to create a new model tailored for a different but often related task, allowing for the reapplication of insights gained from earlier training cycles.

Transfer learning can enhance the efficiency of model development. It proves especially beneficial when data availability for the targeted task is limited. Nevertheless, it’s essential to be aware that this method also has downsides; models using transfer learning to acquire general competencies will likely need additional training with new data to excel in their specific areas of focus.

(See: Fine-tuning)

Validation loss represents a metric that reflects the learning journey of an AI model during training—ideally, lower values are preferred. Researchers monitor this metric closely as a real-time performance indicator, guiding decisions on when to stop training, modify hyperparameters, or investigate potential issues. A primary concern that validation loss helps illuminate is overfitting, where a model memorizes its training data instead of genuinely understanding applicable patterns for new scenarios. Consider validation loss as a tool that distinguishes between a student who comprehends the material and one who merely memorizes previous exam questions—it encapsulates your model’s developmental evolution.

Weights are essential components of AI training, determining the significance (or weight) assigned to various attributes (or input variables) within the training data, ultimately influencing the outputs produced by the AI model.

Essentially, weights are numerical parameters that indicate which elements within a dataset are most relevant for the specified training task. They accomplish this by multiplying with inputs. Training usually begins with randomly assigned weights, but as the process continues, these weights are modified to yield outputs that closely reflect the desired results.

For instance, an AI model aimed at predicting housing prices based on historical real estate data for a specific area might assign weights to attributes such as the number of bedrooms and bathrooms, whether a property is semi-detached or detached, and if it includes parking or a garage.

In the end, the weights assigned reflect the importance of each of these factors in determining a property’s value based on the available dataset.

This article is regularly updated with new information.