Essential AI Terms to Know in 2023

The realm of artificial intelligence is progressing swiftly, bringing forth an assortment of terms to articulate its various ideas. In contemporary tech conferences or product unveilings, terms such as LLMs, RAG, RLHF, and others can perplex even seasoned professionals. This glossary seeks to elucidate these terms by offering clear definitions of AI terminology you may come across—whether you’re delving into these technologies, investing, or staying informed through news outlets like TechCrunch or similar podcasts. We will keep it updated to mirror the ever-changing landscape of AI systems as the industry advances.

---

Artificial general intelligence (AGI) remains a challenging concept, typically denoting AI that surpasses average human abilities across various tasks. OpenAI CEO Sam Altman likens AGI to the “average human one could employ as a coworker.” OpenAI characterizes AGI as “highly autonomous systems that excel in most economically meaningful activities.” Meanwhile, Google DeepMind defines it as “AI that can effectively perform most cognitive tasks at a human-like level.” If this seems ambiguous, you’re in good company; many in the AI research community feel similarly.

An AI agent represents a tool that utilizes AI technologies to automate a wide array of tasks, extending well beyond conventional chatbot functionalities. Tasks managed by AI agents may include expense tracking, travel organization, or even the writing and maintaining of software. As mentioned earlier, interpretations of an “AI agent” can vary among stakeholders, and its foundational framework is continually evolving to meet escalating expectations. Fundamentally, an AI agent operates as an autonomous system that integrates various AI technologies to handle complex, multi-step tasks.

Consider API endpoints as “hidden buttons” within software that other applications can trigger to execute tasks. Developers use these interfaces to create integrations—like enabling one application to retrieve data from another or allowing an AI agent to autonomously interact with third-party services. Many smart home devices and interconnected systems harbor these inconspicuous buttons, frequently overlooked by average users. As AI agents mature, their ability to independently identify and utilize these endpoints is improving, revealing powerful—and sometimes unforeseen—automation capabilities.

When confronted with simple questions, humans typically respond instinctively—like, “which is taller, a giraffe or a cat?” Yet, some questions necessitate breaking down intermediate steps to derive an accurate answer. For example, if a farmer owns 40 heads of chickens and cows combined with 120 legs, one must formulate an equation to determine how many of each there are (20 chickens and 20 cows).

In AI, chain-of-thought reasoning used by large language models involves dissecting a problem into smaller, sequential steps to enhance the accuracy of the final answer. This method may take longer to reach a solution but significantly boosts correctness probabilities, especially in logical or programming scenarios. Reasoning models emerge from conventional large language models and are refined for chain-of-thought techniques, largely via reinforcement learning.

(See: Large language model)

This term is more precise than “AI agent,” emphasizing applications capable of independently executing tasks sequentially to attain desired results. A coding agent is a specific variety focused exclusively on software development. Instead of merely generating code for human review, a coding agent can autonomously write, test, and debug software, tackling the iterative challenges that developers often face. These agents can navigate complete codebases, identify bugs, conduct testing, and apply fixes with minimal human oversight, operating like an exceptionally efficient intern—hardworking and relentless, yet still requiring human direction.

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

Deep learning, a subset of self-improving machine learning, focuses on developing AI algorithms within a layered artificial neural network (ANN) structure. This architecture allows for the recognition of more sophisticated patterns compared to traditional machine learning techniques like linear models or decision trees. Deep learning systems draw inspiration from the complex pathways of neurons in the human brain.

Deep learning models can autonomously identify critical features within data, eliminating the necessity for human engineers to define them. This architecture also enables algorithms to learn from errors, progressively improving outputs through iterative updates. However, deep learning systems usually require substantial amounts of data for optimal performance (often millions of data points) and generally experience extended training periods compared to simpler models—resulting in higher development expenses.

(See: Neural network)

Diffusion refers to the methods employed by various AI models to produce art, music, and text. Derived from physics, diffusion systems systematically “destroy” data structures—like images or melodies—by introducing noise until they become indistinguishable. In physics, diffusion happens spontaneously and irreversibly; for instance, sugar dissolved in coffee cannot revert to its original form. In contrast, AI diffusion systems aim to master a “reverse diffusion” technique to reclaim distorted data, effectively learning to extract information hidden by noise.

Distillation denotes a technique for extracting insights from a large AI model using a “teacher-student” framework. Developers interact with a teacher model and document its outputs, often comparing these results against a predetermined dataset for accuracy. These outputs are then utilized to train the student model, which aims to replicate the teacher’s responses.

This method can generate a more compact, efficient model that retains the effectiveness of its larger counterpart while minimizing distillation loss. It is widely believed that this technique contributed to the creation of GPT-4 Turbo, a faster variant of GPT-4 from OpenAI.

While all AI organizations conduct distillation internally, some might use it to enhance their features compared to leading models. However, extracting insights from a competitor may breach AI API service agreements.

This term indicates the retraining of an AI model to optimize its performance for a more specific task or domain than initially intended—typically by integrating new, task-relevant data.

Many AI startups begin with large language models as the foundation for their products, intending to enhance functionality for particular sectors or tasks by complementing prior training with domain-specific knowledge.

(See: Large language model [LLM])

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

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

Within the AI community, “hallucination” refers to instances where AI models generate inaccuracies—essentially fabricating false information. This poses a significant challenge for AI accuracy.

Hallucinations can result in misleading outputs from generative AI, posing considerable risks—potentially leading to harmful consequences (e.g., unsafe medical advice from a health query).

The issue of AIs producing fictitious information is thought to originate from gaps in their training data. Consequently, there is an increasing push towards developing more specialized and/or vertical AI models—domain-specific AIs that require narrower expertise to bridge knowledge gaps and mitigate misinformation risks.

Inference signifies the execution of an AI model; it includes activating a model to predict or draw conclusions from previously observed data. Notably, inference cannot occur without training; a model must first recognize patterns within a dataset before effectively extrapolating from that training.

Various hardware configurations can handle inference, spanning consumer-grade processors to high-performance GPUs and specialized AI accelerators. However, not all setups execute models with equal efficacy; for instance, exceptionally large models would necessitate significantly more time to generate predictions on a laptop compared to a cloud server equipped with advanced AI chips.

[See: Training]

Large language models, or LLMs, are the AI frameworks powering popular AI assistants like 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 directly responds to your queries or utilizes various available tools, such as web browsers or code interpreters.

LLMs comprise deep neural networks with billions of numerical parameters (or weights, see below) that identify relationships among words and phrases to construct a representation of language—essentially a sophisticated map of language.

These models are built by analyzing patterns found in extensive quantities of documents, articles, and transcripts. When you provide a prompt to an LLM, the model generates the most plausible response related to that prompt.

(See: Neural network)

Memory cache signifies an essential process enhancing inference (how AI formulates responses to user prompts). Essentially, caching acts as an optimization technique aimed at improving inference performance. AI operates through intricate mathematical computations, where each calculation consumes resources. Caching diminishes the number of computations a model needs to perform by retaining certain calculations for future interactions. Several forms of memory caching exist, with KV (key-value) caching being among the most common. KV caching is utilized in transformer-based models, boosting efficiency and delivering quicker results by minimizing the time (and computational effort) required to generate responses to user prompts.

(See: Inference)

Model Context Protocol (MCP) is an open standard designed to streamline AI models’ interaction with external tools and data—such as your files, databases, or applications like Slack and Google Drive—without requiring developers to create individual connectors for each integration. Consider it a USB-C port for AI. Launched by Anthropic in 2024 and subsequently supported by the Linux Foundation, MCP has swiftly gained traction among OpenAI, Google, and Microsoft, rapidly becoming one of the most widely adopted standards in the current 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—similar to consulting the entire team for every inquiry—the MoE model employs a built-in “router” that selects the most pertinent specialists for the task. This structure allows for the creation of exceptionally large models while maintaining efficiency and cost-effectiveness, as only a fraction of the network is active at any time. Mistral AI’s Mixtral model is a notable instance; recent GPT models from OpenAI are rumored to utilize some variant of this approach, although the company has yet to officially confirm it.

(See: Neural network, Deep learning)

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

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

(See: Large language model [LLM])

Open source pertains to software—or increasingly, AI models—where the underlying code is publicly available, 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 well-known benchmark among operating systems. Open-source principles empower researchers, developers, and organizations globally to build upon each other’s work, promoting innovation and facilitating independent safety evaluations that closed systems frequently lack. Closed source means that the code remains proprietary; users can engage with the product but cannot scrutinize its internal workings, a characteristic often attributed to OpenAI’s GPT models—which has triggered considerable debate within the AI community.

Parallelization entails performing multiple tasks simultaneously rather than sequentially—akin to employing ten individuals to address different aspects of a project simultaneously rather than having one person complete each task in succession. In AI, parallelization is crucial for both training and inference; modern GPUs are specifically designed to handle thousands of calculations concurrently, forming the backbone of their role in the industry. As AI systems become more intricate and models expand, the ability to parallelize tasks across multiple chips and machines has emerged as a vital factor in determining how swiftly and efficiently models can be developed and deployed. Research into improved parallelization techniques has transformed into a specialized discipline.

RAMageddon is a humorous term reflecting a significant trend influencing the tech landscape: the ongoing shortage of random access memory (RAM) chips, which are essential for nearly all technology products. As the AI sector has surged, major tech firms and AI laboratories have been acquiring RAM at unmatched levels to bolster their data centers, resulting in decreased availability in the wider market. This supply crunch is inflating RAM prices.

This shortage impacts various sectors, such as gaming (where companies are compelled to hike console prices due to challenges in sourcing 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 adequate RAM for their data operations). The escalating costs are unlikely to stabilize until this projected shortage alleviates, but unfortunately, indications 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 start to enhance their capabilities independently, leading to significant advancements in autonomy and functionality. Some narratives liken this phenomenon 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 for engineers attempting to develop such models. Numerous emerging AI startups concentrate on creating 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 where a system learns through trial and error, receiving rewards for correct actions—similar to training a pet with treats, though in this scenario, the “pet” is a neural network, and the “treat” is a mathematical signal indicating success. Unlike supervised learning, which involves training a model on a fixed dataset of labeled examples, reinforcement learning allows a model to engage with its environment, take actions, and iteratively refine its behavior based on the feedback received. This method has proven particularly effective for training AI to compete in games, control robotics, and, more recently, improve the reasoning skills of large language models. Approaches like reinforcement learning from human feedback (RLHF) have become integral in how leading AI labs enhance their models for better performance, precision, and safety.

In the interactions between humans and machines, noteworthy challenges arise—humans communicate in natural language, while AI systems operate through complex algorithmic processes based on data. Tokens act as the bridge 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 formed through a method known as tokenization, which divides raw text into manageable parts that a language model can comprehend, much like how a compiler converts human language into binary for computers. In business contexts, tokens also impact costs—many AI companies charge for LLM usage on a per-token basis, indicating that increased utilization correlates with higher expenses.

As such, tokens are small text units—typically fragments of words rather than complete phrases—that AI language models analyze prior to processing; they can be compared to “words” regarding AI workloads. Throughput denotes the amount that can be processed within a specified timeframe, making token throughput a measure of the volume of AI work a system can handle concurrently. High token throughput is a critical objective for AI infrastructure teams, as it influences how many users a model can assist simultaneously and the speed with which they receive responses. AI researcher Andrej Karpathy has voiced concerns about 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 priority in the field.

Training machine learning AIs encompasses a procedure known as training. This primarily involves supplying data to allow the model to discern patterns and generate desired outputs. It represents 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 considerable resources due to the vast amounts of inputs required, which have been steadily increasing—especially with the importance of hybrid strategies, like merging rules-based AIs with targeted data, to manage costs without initiating the process from scratch.

[See: Inference]

This method employs a previously trained AI model as a baseline for developing a new model aimed at a different but typically related task, allowing the reapplication of insights obtained from earlier training cycles.

Transfer learning can enhance model development efficiency. It proves especially beneficial when the available data for the target task is scarce. However, it’s important to recognize that this approach has limitations; models using transfer learning to acquire generalized skills will likely need 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 preferred. Researchers closely monitor this measure as a real-time performance indicator, aiding decisions regarding when to halt training, adjust hyperparameters, or explore potential issues. A significant concern that validation loss helps to identify is overfitting, where a model memorizes its training data rather than genuinely understanding applicable patterns for new situations. Think of validation loss as a tool that differentiates between a student who grasps the material and one who merely memorizes last year’s exam—it encapsulates the developmental journey of your model.

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 indicating which elements within a dataset are most critical for the specific training task. They achieve this effect through multiplication with inputs. Training typically commences with randomly initialized weights, but as the process unfolds, these weights are refined as the model strives to produce outputs closely aligned with the desired outcomes.

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

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

This article is regularly updated with new information.