Essential AI Terminology to Understand in 2023

The domain of artificial intelligence is advancing swiftly, bringing forth a myriad of terms to conceptualize its ideas. At contemporary tech conferences and product unveilings, acronyms such as LLMs, RAG, RLHF, and various others can bewilder even seasoned professionals. This glossary is designed to demystify these terms by offering clear definitions of AI terminology that you may come across—whether you’re delving into these technologies, investing, or staying updated with media like TechCrunch or relevant podcasts. We’ll ensure it stays relevant to reflect the ever-evolving landscape of AI systems as the field progresses.

---

Artificial general intelligence (AGI) is a multifaceted concept that generally signifies AI that exceeds human capabilities over a diverse array of tasks. OpenAI’s CEO Sam Altman likens AGI to the “average human you might employ as a coworker.” OpenAI labels AGI as “highly autonomous systems that excel in most economically significant activities.” Meanwhile, Google DeepMind characterizes it as “AI capable of competently performing most cognitive tasks at a human-like level.” If this characterization feels ambiguous, you’re not alone; many within the AI research community express similar sentiments.

An AI agent is a tool that harnesses AI technologies to automate a wide range of tasks, extending well beyond conventional chatbot functionalities. Duties may include monitoring expenses, planning travel, or even writing and maintaining software. As noted earlier, the definition of an “AI agent” can vary among stakeholders, and its foundational structure is continually adapting to meet rising expectations. At its core, an AI agent operates as an autonomous system that amalgamates various AI technologies to oversee intricate, multi-step tasks.

Consider API endpoints as “hidden buttons” embedded in software, which other applications can initiate to perform functions. Developers leverage these interfaces to craft integrations—such as enabling one application to extract data from another or allowing an AI agent to autonomously interact with third-party services. Many smart home devices and interconnected systems include these discreet buttons, often overlooked by the average user. As AI agents evolve, their ability to recognize and utilize these endpoints improves, revealing powerful—and occasionally unexpected—automation capabilities.

When presented with straightforward inquiries, humans typically respond instinctively—such as asking, “which is taller, a giraffe or a cat?” Yet, some queries necessitate breaking down intermediate steps to yield accurate answers. For instance, if a farmer possesses 40 heads of chickens and cows together with 120 legs, one must establish an equation to determine how many of each exist (20 chickens and 20 cows).

In AI, chain-of-thought reasoning employed by large language models involves dissecting a problem into smaller, sequential steps to enhance the accuracy of the final answer. Although this method may take longer to yield a solution, it significantly boosts the chances of correctness, particularly in logical or programming scenarios. Reasoning models originate from classical large language models and are refined for chain-of-thought processes, largely through reinforcement learning.

(See: Large language model)

This term is more specific than “AI agent,” emphasizing applications capable of autonomously executing tasks in sequence to achieve desired outcomes. A coding agent is a specialized variant focusing exclusively on software development. Rather than merely generating code for human review, a coding agent can independently write, test, and debug software, tackling the iterative challenges developers frequently encounter. These agents can navigate entire codebases, identify bugs, run tests, and implement fixes with minimal human oversight, operating much like a highly diligent intern—hardworking and tireless, yet still requiring human direction.

Though somewhat broad, compute primarily refers to the computational power that fuels AI models. This capability underpins the AI sector, enabling the training and deployment of robust models. It often pertains to the types of hardware that provide this computational capacity—such as GPUs, CPUs, TPUs, and other vital infrastructure critical to today’s AI ecosystem.

Deep learning, a subset of self-improving machine learning, focuses on creating AI algorithms within a layered artificial neural network (ANN) framework. This design allows for the recognition of more complex patterns compared to traditional machine learning approaches like linear models or decision trees. Deep learning systems take cues from the intricate pathways of neurons in the human brain.

Deep learning models can autonomously identify significant features within data, eliminating the requirement for human engineers to specify them. This architecture also allows algorithms to learn from errors, gradually fine-tuning outputs via iterative updates. However, deep learning systems generally necessitate vast amounts of data for optimal performance (often millions of data points) and usually experience longer training periods compared to simpler models—resulting in higher development costs.

(See: Neural network)

Diffusion pertains to methods that various AI models employ to generate art, music, and text. Taken from physics, diffusion systems systematically “destroy” data structures—like images or melodies—by introducing noise until they become indistinguishable. In physics, diffusion occurs spontaneously and irreversibly; for instance, sugar dissolving in coffee can’t revert to its original state. Conversely, AI diffusion systems attempt to master a “reverse diffusion” technique to recover distorted data, effectively learning to extract information obscured by noise.

Distillation refers to a method for extracting insights from a large AI model using a “teacher-student” framework. Developers engage with a teacher model and document its outputs, frequently comparing these results to a predetermined dataset for accuracy. These outputs are subsequently used to train a student model, aiming to replicate the teacher’s responses.

This method can yield a more compact, efficient model that retains the effectiveness of its larger counterpart while minimizing distillation loss. It’s widely believed that this technique played a role in the development of GPT-4 Turbo, a faster iteration of GPT-4 from OpenAI.

While all AI organizations perform distillation internally, some may exploit it to enhance their features compared to dominant models. However, extracting insights from a competitor may breach AI API service agreements.

This term signifies the retraining of an AI model to refine its performance for a more specific task or domain than it was originally intended—typically through the integration of new, task-related data.

Many AI startups kick off with large language models as the foundation for their products, aiming to enhance functionality for specific sectors or tasks by supplementing prior training with domain-specific knowledge.

(See: Large language model [LLM])

A GAN, or Generative Adversarial Network, is a machine learning framework that has greatly advanced generative AI, particularly in creating realistic outputs—including deepfake technologies. GANs comprise two neural networks; one generates outputs based on its training, while the other assesses them.

These two networks engage in a competitive dynamic; the generator aims to produce outputs that can mislead the discriminator, while the discriminator strives to distinguish genuine data from fabricated content. This structured rivalry can result in realistic AI outputs with minimal human intervention. Nevertheless, GANs tend to excel only under specific conditions (e.g., generating authentic images or videos) rather than being universally applicable.

Within the AI community, “hallucination” refers to instances when AI models create inaccuracies—essentially fabricating false information. This presents a significant challenge to AI accuracy.

Hallucinations can yield misleading outputs from generative AI, posing serious risks—potentially resulting in harmful outcomes (e.g., unsafe medical advice from a health inquiry).

The phenomenon of AIs producing fictional information is believed to arise from gaps in their training data. Consequently, there is an increasing demand for developing more specialized and/or vertical AI models—domain-specific AIs that necessitate narrower expertise to bridge knowledge gaps and minimize misinformation risks.

Inference refers to the execution of an AI model, activating the model to predict or deduce conclusions based on previously observed data. Notably, inference cannot occur without prior training; a model must first learn patterns within a dataset before effectively extrapolating from that training.

Various hardware configurations can facilitate inference, ranging from consumer-grade processors to high-performance GPUs and specialized AI accelerators. However, not all setups execute models with equivalent efficiency; for example, exceedingly large models would take significantly longer to generate predictions on a laptop compared to a cloud server equipped with advanced AI chips.

[See: Training]

Large language models, or LLMs, serve as the foundational AI frameworks behind popular AI assistants including ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, and Mistral’s Le Chat. When engaging with an AI assistant, you are utilizing a large language model that responds directly to your queries or employs various available tools, such as web browsers or code interpreters.

LLMs are composed of deep neural networks with billions of numerical parameters (or weights) that discern relationships among words and phrases to construct a representation of language—essentially a sophisticated mapping of language.

These models are developed by analyzing patterns captured in extensive volumes of documents, articles, and transcripts. When you provide a prompt to an LLM, the model generates the most probable response related to that prompt.

(See: Neural network)

Memory cache denotes a crucial process that enhances inference (how AI formulates responses to user prompts). Essentially, caching serves as an optimization technique aimed at enhancing inference performance. AI operates through complex mathematical computations, each calculation consuming resources. Caching diminishes the number of computations a model needs to execute by retaining certain calculations for future interactions. Various types of memory caching exist, with KV (key-value) caching being among the most prevalent. KV caching is utilized in transformer-based models, improving efficiency and providing quicker results by lessening the time (and computational effort) needed to generate responses to user prompts.

(See: Inference)

Model Context Protocol (MCP) is an open standard designed to simplify AI models’ interaction with external tools and data—such as your files, databases, or applications like Slack and Google Drive—without necessitating developers to create individual connectors for each integration. Think of it as a USB-C port for AI. Introduced by Anthropic in 2024 and later supported by the Linux Foundation, MCP has swiftly gained popularity among OpenAI, Google, and Microsoft, becoming one of the most widely embraced 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 of them for specific tasks. Rather than routing each request through the entire model—akin to consulting the entire team for every question—the MoE model employs an intrinsic “router” that selects the most relevant specialists for the job. This structure enables the creation of exceedingly large models while maintaining efficiency and cost-effectiveness, as only a portion of the network remains active at any given moment. Mistral AI’s Mixtral model is a noteworthy example; recent GPT models from OpenAI are rumored to utilize a variation of this method, although the company has yet to confirm it officially.

(See: Neural network, Deep learning)

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

The concept of modeling data-processing algorithms after the intricately interlinked circuits of the human brain has existed since the 1940s; however, it was the recent surge in graphical processing units (GPUs)—primarily driven by the gaming sector—that truly unlocked the potential of this theory. GPUs are specifically designed for training algorithms with multiple layers, enabling neural network-based AI systems to excel across various fields, such as voice recognition, autonomous driving, and drug discovery.

(See: Large language model [LLM])

Open source pertains to software—or increasingly, AI models—where the underlying code is accessible to the general public, allowing anyone to utilize, inspect, or modify it. In the AI domain, Meta’s Llama series of models stands out as a prominent example; Linux serves as a widely recognized benchmark among operating systems. Open-source principles empower researchers, developers, and organizations globally to build upon one another’s work, fostering innovation and enabling independent safety assessments that closed systems often lack. Closed source indicates that the code remains proprietary; users can engage with the product but cannot scrutinize its internal mechanics, a characteristic frequently associated with OpenAI’s GPT models—which has generated considerable debate within the AI community.

Parallelization involves executing multiple tasks simultaneously instead of sequentially—akin to employing ten individuals to address different aspects of a project concurrently, rather than having one person complete each task in succession. In AI, parallelization is critical for both training and inference; modern GPUs are specifically designed to manage thousands of calculations at once, forming the backbone of their role in the industry. As AI systems become increasingly complex and models expand, the ability to parallelize tasks across multiple chips and machines has become a vital factor in determining how swiftly and efficiently models can be developed and deployed. Research into enhanced parallelization techniques has matured into a specialized discipline.

RAMageddon is a humorous term reflecting a significant trend affecting the tech landscape: the ongoing shortage of random access memory (RAM) chips, which are vital for nearly all technology products. As the AI sector surges, major tech companies and AI laboratories have been acquiring RAM at unprecedented levels to fortify their data centers, resulting in diminished availability in the broader market. This supply crunch is escalating RAM prices.

This shortage impacts various sectors, including gaming (where companies are compelled to raise console prices due to challenges in procuring memory chips for their devices), consumer electronics (where memory constraints could lead to the lowest smartphone shipments in over a decade), and general corporate computing (as businesses strive to secure sufficient RAM for their data operations). The rising costs are unlikely to stabilize until this projected shortage eases, but unfortunately, signs of a swift resolution are scarce.

Similar to AGI, recursive self-improvement suggests a potential breakthrough in AI development, enabling minimal human oversight. In this context, AI models begin to enhance their capabilities autonomously, leading to significant advancements in autonomy and functionality. Some narratives liken this scenario to a singularity moment—when AI models become immune to external influence. However, recursive self-improvement raises a fundamental question: can an AI model create its successor? This inquiry simplifies the challenges engineers face in developing such models. Numerous emerging AI startups aim to create recursively self-improving models while often downplaying the potentially catastrophic implications, framing recursive self-improvement merely as the next frontier in research.

Reinforcement learning is an AI training approach in which a system learns through trial and error, receiving rewards for correct actions—similar to instructing a pet with treats, although in this case, the “pet” is a neural network, and the “treat” is a mathematical signal denoting success. Unlike supervised learning, which encompasses training a model on a fixed dataset of labeled examples, reinforcement learning enables a model to interact with its environment, undertake actions, and iteratively refine its behavior based on received feedback. This approach has proven particularly effective for training AI in games, controlling robotics, and, more recently, enhancing the reasoning capabilities of large language models. Techniques such as reinforcement learning from human feedback (RLHF) have become essential in how leading AI labs refine their models for improved performance, precision, and safety.

In the interactions between humans and machines, notable challenges arise—humans communicate in natural language, while AI systems operate through complex algorithmic processes based on data. Tokens serve as the link between these two realms: they represent the fundamental units of human-AI interaction, encapsulating distinct segments of data processed or generated by an LLM. Tokens are produced through a process called tokenization, which divides raw text into manageable parts that a language model can grasp, similar to how a compiler translates human language into binary for computers. In business contexts, tokens also influence costs—many AI companies charge for LLM usage on a per-token basis, indicating that increased utilization correlates with higher expenses.

Thus, tokens are small text units—typically fragments of words rather than complete phrases—that AI language models analyze prior to processing; they can be equated to “words” regarding AI workloads. Throughput refers to the amount that can be processed within a specified timeframe, making token throughput a measure of the volume of AI work a system can manage simultaneously. High token throughput is a critical goal for AI infrastructure teams since it impacts how many users a model can assist concurrently and the speed at which they receive responses. AI researcher Andrej Karpathy has voiced concerns regarding his AI subscriptions being underutilized—reflecting on his experiences as a graduate student when costly computing hardware was used inefficiently—underscoring why optimizing token throughput has become an urgent focus in the field.

Training machine learning AIs encompasses a process known as training. This primarily involves supplying data to enable the model to recognize patterns and produce desired outputs. It signifies the system’s response to the characteristics of the data, aligning its outputs toward a desired outcome—whether that entails identifying images of cats or crafting a haiku in response to a prompt.

Training can demand substantial resources due to the vast amounts of inputs required, which have been consistently increasing—especially with the significance of hybrid strategies, like merging rules-based AIs with targeted data, to manage costs without starting the process from scratch.

[See: Inference]

This technique employs a previously trained AI model as a foundation for developing a new model directed at a different but often related task, allowing the reapplication of insights gained from earlier training cycles.

Transfer learning can enhance model development efficiency. It proves especially advantageous when the data available for the target task is limited. However, it’s essential to recognize that this approach also has constraints; models leveraging transfer learning to acquire generalized skills will likely require additional training with new data to excel in their specific focus areas.

(See: Fine tuning)

Validation loss serves as a metric reflecting an AI model’s learning progress throughout training—ideally, lower values are more desirable. Researchers monitor this measure closely as a real-time performance indicator, aiding decisions about when to halt training, adjust hyperparameters, or dig into potential issues. A significant concern that validation loss helps to identify is overfitting, where a model memorizes its training data rather than genuinely comprehending applicable patterns for new situations. Think of validation loss as a tool that distinguishes between a student who grasps the material and one who merely memorizes previous exam content—it encapsulates your model’s developmental journey.

Weights are fundamental to AI training, determining the significance (or weight) assigned to various features (or input variables) within the training data, ultimately influencing the AI model’s outputs.

In simpler terms, weights are numerical parameters that indicate which elements within a dataset are most critical for the specific training task. They accomplish this through multiplication with inputs. Training typically commences with randomly assigned weights, but as the process unfolds, these weights are refined to yield outputs that closely align with desired outcomes.

For instance, an AI model designed to predict housing prices based on historical real estate data for a specific locale might assign weights to variables like the number of bedrooms and bathrooms, whether a property is semi-detached or detached, and if it features parking or a garage.

Ultimately, the weights assigned reflect how significantly each of these factors influences a property’s value according to the available dataset.

This article is regularly updated with new information.