Essential AI Terms to Understand in 2023

The field of artificial intelligence is rapidly evolving, bringing forth a variety of new terminology designed to clarify its intricacies. At recent tech gatherings and product unveilings, terms like LLMs, RAG, RLHF, and more can confuse even the most seasoned professionals. This glossary is designed to simplify these terms, providing straightforward definitions for the AI terminology you may encounter—whether you’re exploring these technologies, investing in them, or keeping up with updates from sources like TechCrunch or relevant podcasts. We will continuously update it to reflect the fast-paced AI environment.

---

Artificial general intelligence (AGI) represents a high-level concept generally describing AIs capable of outperforming human abilities across multiple tasks. OpenAI’s CEO, Sam Altman, compares AGI to “the average employee you might hire.” According to OpenAI, AGI refers to “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 thoughts.

An AI agent is a system that utilizes AI technologies to perform a broad range of tasks, significantly exceeding the capabilities of a standard chatbot. This can involve tasks such as budget management, travel planning, and coding, among others. Interpretations of “AI agent” can differ among various stakeholders, and its underlying framework constantly adapts to meet rising expectations. In essence, an AI agent functions as an autonomous entity integrating multiple AI technologies to tackle complex, multi-step tasks.

Think of API endpoints as “invisible buttons” in software, activated by other applications to perform various functions. Developers utilize these interfaces for seamless integration—allowing one application to retrieve data from another or enabling an AI agent to perform autonomously with third-party services. Many smart home devices and interconnected systems utilize these subtle controls, often without the average user being aware. As AI agents advance, their ability to discover and utilize these endpoints will expand, creating remarkable—and sometimes unexpected—automation capabilities.

When faced with straightforward questions, humans often respond intuitively—like asking, “Which is taller, a giraffe or a cat?” However, some questions require breaking down intermediate steps to arrive at the correct answers. For instance, if a farmer has 40 animals (chickens and cows) accounting for 120 legs, one must calculate to determine the type of each animal (20 chickens and 20 cows).

In AI, chain-of-thought reasoning employed by large language models involves breaking problems down into smaller, sequential steps to improve the accuracy of the final answer. While this method may take additional time, it significantly increases correctness, especially in logical or programming tasks. Reasoning models evolve from standard 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 compared to “AI agent,” focusing on applications capable of autonomously performing tasks in sequence to achieve defined goals. A coding agent is a specialized type of AI primarily aimed at software development. Rather than only producing code for human validation, a coding agent can autonomously write, test, and debug software, addressing the iterative challenges faced by developers. These agents can navigate entire codebases, identify bugs, run tests, and apply fixes with little human intervention, resembling diligent interns—industrious and persistent, yet still requiring human guidance.

While somewhat broad, compute generally relates to the computational power that enables AI models. This capacity is vital in the AI landscape, facilitating the training and execution of complex models. It typically pertains to the types of hardware providing this computational strength—such as GPUs, CPUs, TPUs, and other infrastructures essential to the modern AI ecosystem.

Deep learning, a subset of self-improving machine learning, focuses on developing AI algorithms that use a layered artificial neural network (ANN) structure. This technique allows for the identification of more intricate patterns compared to traditional training methods like linear models or decision trees. Deep learning systems are inspired by the complex pathways of neurons in the human brain.

Deep learning models can autonomously identify significant features within data, reducing the need for human engineers to pinpoint them. This framework also allows algorithms to learn from mistakes, continuously refining outputs through iterative adjustments. However, deep learning systems typically require large datasets to operate optimally (often millions of data points) and generally demand longer training times compared to simpler models, resulting in increased development costs.

(See: Neural network)

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

Distillation describes a process for extracting insights from a large AI model using a “teacher-student” framework. Developers interact with a teacher model and carefully document its outputs, often benchmarking these against a defined dataset for accuracy. These results train a student model that mimics the teacher’s responses.

This approach can result in a more compact, effective model that retains the abilities of its larger counterpart while minimizing distillation loss. It is widely believed that this process contributed to the creation of GPT-4 Turbo, a faster version of GPT-4 from OpenAI.

While all AI organizations conduct distillation internally, some may leverage it to enhance their features relative to existing models. However, extracting insights from a competitor could violate AI API service agreements.

This term indicates the retraining of an AI model to enhance its performance for a specific task or domain from which it was initially developed—usually accomplished by adding data relevant to the task.

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

(See: Large language model [LLM])

A GAN, or Generative Adversarial Network, functions as a machine learning framework critical in advancing generative AI, especially for producing realistic outputs—including deepfakes. GANs consist of two neural networks: one that generates outputs based on its training, and another that evaluates them.

These two networks maintain a competitive relationship; the generator strives to produce outputs that can deceive the discriminator, while the discriminator tries to distinguish real data from generated content. This structured competition can yield lifelike AI-generated outputs with minimal human involvement. However, GANs tend to perform best under specific conditions (e.g., when creating authentic images or videos) rather than universally.

Within the AI community, “hallucination” refers to instances where AI models produce inaccuracies—essentially generating false information. This represents a significant challenge to AI accuracy.

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

The phenomenon of AIs generating fictitious data is believed to stem from gaps in their training data. Hence, there is a growing demand for more specialized and/or vertical AI models—domain-specific AIs that require narrowed expertise to fill knowledge gaps and mitigate misinformation risks.

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

Various hardware configurations can support inference, ranging from consumer-grade processors to high-end GPUs and specialized AI accelerators. However, not all setups process models with the same efficiency; for example, exceptionally large models would take substantially longer to generate predictions on a laptop compared to a cloud server equipped with cutting-edge AI chips.

[See: Training]

Large language models, or LLMs, serve as the foundation for widely used AI assistants, including ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, and Mistral’s Le Chat. When you interact with an AI assistant, you’re engaging with a large language model that responds to your inquiries or makes use of various available tools, such as web browsers or code interpreters.

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

These models are trained by investigating patterns identified within extensive datasets of documents, articles, and transcripts. When you prompt an LLM, the model generates the most likely response relevant to that input.

(See: Neural network)

Memory cache is an essential process that optimizes inference (how AI formulates responses to user prompts). Essentially, caching serves as an optimization strategy aimed at boosting inference performance. AI operations demand complex mathematical calculations, each straining resources. Caching reduces the number of computations a model needs to perform by storing certain calculations for future use. Various types of memory caching exist, with KV (key-value) caching being among the most common. KV caching is utilized in transformer-based models, enhancing efficiency and yielding quicker results by minimizing the time (and computational effort) needed to respond to user inquiries.

(See: Inference)

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

The Mixture of Experts architecture segments a neural network into several smaller, specialized sub-networks, or “experts,” activating only a selected subset for designated tasks. Instead of processing every request through the entire model—similar to consulting a full team for each task—the MoE model employs an internal “router” to select the most pertinent specialists for a particular job. This approach facilitates the creation of exceedingly large models while maintaining efficiency and cost-effectiveness, as only a limited segment of the network functions at any given time. Mistral AI’s Mixtral model exemplifies this; recent GPT models from OpenAI are rumored to implement a variant of this methodology, although the company has not publicly affirmed this.

(See: Neural network, Deep learning)

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

The concept of modeling data-processing algorithms after the intricately interconnected circuits of the human brain has existed since the 1940s; however, the recent surge of graphical processing units (GPUs)—primarily driven by the gaming sector—has truly unlocked the potential of this theory. GPUs are specifically designed for training algorithms with multiple layers, allowing AI systems based on neural networks to shine in various domains, including voice recognition, autonomous navigation, and drug discovery.

(See: Large language model [LLM])

Open source refers to software—or increasingly, AI models—where the source code is publicly accessible, allowing anyone to use, examine, or modify it. In the AI domain, Meta’s Llama series of models exemplifies a significant instance, while Linux is recognized as a benchmark among operating systems. Open-source practices empower researchers, developers, and organizations globally to benefit from each other’s contributions, fostering innovation and enabling independent safety assessments often absent in closed systems. Closed source indicates that the code remains proprietary; users can utilize the product but cannot inspect its inner mechanisms, a characteristic often associated with OpenAI’s GPT models—which has sparked considerable debate within the AI community.

Parallelization involves executing multiple tasks concurrently rather than sequentially—akin to employing ten individuals to handle different aspects of a project simultaneously instead of relying on one person to tackle each task in order. In AI, parallelization is crucial for both training and inference; modern GPUs are specifically engineered to manage thousands of calculations at once, which is fundamental to their role in the field. As AI systems become more intricate and models expand, the ability to parallelize tasks across multiple chips and machines has become vital in determining the speed and efficiency of model development and deployment. Research into advanced parallelization techniques has emerged as a focused area of investigation.

RAMageddon is a humorous term that captures a notable trend affecting the tech industry: the ongoing scarcity of random access memory (RAM) chips, integral to nearly every technology product. As the AI landscape thrives, major tech firms and AI labs have been acquiring RAM at an unprecedented rate to enhance their data centers, leading to reduced availability in the general market. This supply shortage is driving up RAM prices.

This scarcity impacts various sectors, including gaming (where companies face elevated console prices due to challenges in sourcing memory chips for their devices), consumer electronics (where memory limitations could result in the lowest smartphone shipments in over a decade), and overall corporate computing (as businesses strive to secure adequate RAM for their data operations). Price increases are unlikely to stabilize until this anticipated shortage is resolved; however, regrettably, a quick resolution is not in sight.

Similar to AGI, recursive self-improvement refers to a potential breakthrough in AI development that allows for minimal human oversight. In this scenario, AI models begin to autonomously enhance their capabilities, resulting in marked advancements in autonomy and performance. Some narratives liken this phenomenon to a singularity moment—when AI models become independent of external influence. However, recursive self-improvement raises an essential question: can an AI model create its successor? This dilemma underscores the challenges engineers face in creating such systems. Many emerging AI startups are attempting to develop models capable of recursive self-improvement while often downplaying potential catastrophic consequences, presenting recursive self-improvement merely as the next research frontier.

Reinforcement learning is a training approach for AI systems where a model learns through trial and error, receiving rewards for favorable actions—similar to training 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, whereas reinforcement learning enables a model to interact with its environment, execute actions, and iteratively refine its behavior based on feedback received. This technique has proven particularly effective in training AI for games, robotics, and, more recently, enhancing reasoning skills in large language models. Strategies like reinforcement learning from human feedback (RLHF) have become integral to how leading AI labs improve their models for better performance, accuracy, and safety.

In the interaction between humans and machines, significant challenges arise—humans communicate in natural language, while AI systems operate through complex algorithmic processes grounded in data. Tokens serve as the bridge connecting these two worlds: they represent the fundamental units of human-AI interaction, embodying discrete segments of data processed or generated by an LLM. Tokens are formed through tokenization, which disaggregates raw text into manageable components that a language model can comprehend, similar to how a compiler translates human language into binary for computers. In commercial settings, tokens also carry cost implications—many AI companies bill for LLM usage on a per-token basis, meaning that higher consumption results in increased costs.

Thus, tokens are small textual units—often fragments of words rather than full phrases—that AI language models analyze before processing; they can be likened to “words” in AI tasks. Throughput indicates the amount that can be processed within a specified timeframe, with token throughput acting as a measure of the extent of AI work a system can handle simultaneously. High token throughput is a crucial objective for AI infrastructure teams, as it influences how many users a model can support concurrently and the speed at which they receive responses. AI researcher Andrej Karpathy has expressed concerns regarding his AI subscriptions being underutilized—reflecting on his experiences as a graduate student when costly computing hardware was ineffectively employed—highlighting why optimizing token throughput has become an urgent priority in the industry.

Training machine learning AIs refers to a process known as training. This involves providing data to the model to enable it to identify patterns and generate desired outputs. It signifies the system’s response to the characteristics of the data, steering its outputs toward a particular objective—whether that involves classifying images of cats or crafting a haiku in response to a prompt.

Training can require substantial resources due to the immense amounts of input needed, which have consistently increased—particularly with the notion of hybrid strategies, such as combining rule-based AIs with curated data, to manage expenses without restarting the training process.

[See: Inference]

This methodology employs a previously trained AI model as a base for generating a new model aimed at a different but often related task, facilitating the reapplication of insights gained from earlier training cycles.

Transfer learning can enhance model development efficiency. It proves particularly advantageous when the data available for the targeted task is limited. However, it’s essential to recognize that this approach also has downsides; models using transfer learning to acquire general capabilities will likely necessitate further training with new data to excel in their specific focus areas.

(See: Fine-tuning)

Validation loss represents a metric reflecting the learning trajectory of an AI model during training—ideally, lower values are preferable. Researchers closely track this metric as a real-time performance indicator, guiding decisions on when to cease training, adjust 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 contexts. Think of validation loss as a tool that distinguishes between a student who comprehends the material and one who merely memorizes previous exam content—it encapsulates your model’s development journey.

Weights are fundamental components of AI training, determining the importance (or weight) assigned to various features (or input variables) within the training data, ultimately affecting the outputs produced by the AI model.

In essence, weights are numerical parameters that signify which elements within a dataset are most relevant for the specified training task. They achieve this through multiplication with inputs. Training usually starts with randomly assigned weights, but as the process unfolds, these weights are fine-tuned to yield outputs closely aligned with the desired outcomes.

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

Ultimately, the weights assigned reflect the significance of each of these factors in determining a property’s value according to the available dataset.

This article is regularly updated with new information.