Essential AI Concepts to Understand in 2023

The field of artificial intelligence is evolving swiftly, introducing a variety of new terms designed to elucidate its intricacies. Recent technology events and product launches have featured terms such as LLMs, RAG, RLHF, and others, leaving even seasoned professionals scratching their heads. This glossary aims to demystify these terms by providing straightforward definitions of the AI terminology you may encounter—whether you are delving into these technologies, investing in them, or following updates from sources like TechCrunch or relevant podcasts. We will keep this glossary updated to capture the continuously changing landscape of AI.

---

Artificial general intelligence (AGI) generally refers to AIs that surpass human abilities in a wide array of 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,” while Google DeepMind describes it as “AI capable of performing most cognitive tasks at a human level.” If this definition feels somewhat ambiguous, you’re not alone; many in the AI research community feel similarly.

An AI agent is a system that utilizes AI technologies to execute a diverse range of tasks, extending beyond mere chatbot functions. These tasks may encompass budget management, trip planning, and code generation. The interpretation of “AI agent” can differ among various stakeholders, and its foundational architecture is consistently evolving to satisfy ever-higher demands. Essentially, an AI agent acts as an independent entity that combines multiple AI technologies to tackle complex, multi-faceted tasks.

API endpoints can be imagined as “invisible buttons” in software that are activated by other applications to perform different actions. Developers leverage these interfaces for seamless integration, allowing one application to pull data from another or enabling an AI agent to function autonomously with external services. Numerous smart home devices and interconnected systems utilize these hidden controls, often without the average user realizing it. As AI agents advance, their ability to discover and interact with these endpoints will expand, leading to remarkable—and sometimes unexpected—automation capabilities.

When faced with simple queries, humans typically respond instinctively—like when asked, “Which is taller, a giraffe or a cat?” However, certain questions necessitate breaking down intermediary steps to arrive at the correct answer. For instance, if a farmer has 40 animals (chickens and cows) totaling 120 legs, calculations are required to determine the number of each type of animal (20 chickens and 20 cows).

In AI, chain-of-thought reasoning used by large language models involves dissecting problems into smaller, sequential steps to improve the accuracy of the final answer. Although this method may require more time, it significantly enhances precision, particularly in logical or programming tasks. Reasoning models indicate a shift from conventional 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 that can autonomously execute tasks to achieve defined goals. A coding agent serves as a specific example of an AI tailored for software development. Instead of solely generating code for human evaluation, a coding agent can perform, test, and debug software independently, addressing the iterative challenges developers encounter. These agents can navigate entire codebases, identify bugs, run tests, and apply corrections with minimal human oversight, akin to diligent interns—hardworking and persistent, yet in need of some direction.

Though somewhat broad, compute generally relates to the computational capabilities underlying AI models. This power is critical in the AI domain, facilitating the training and operation of complex models. It often pertains to the types of hardware that provide this computational strength—like GPUs, CPUs, TPUs, and other infrastructure essential to today’s AI landscape.

Deep learning, a subdivision of self-improving machine learning, focuses on developing AI algorithms that employ a layered artificial neural network (ANN) structure. This approach enables the recognition of more intricate patterns than traditional training methods such as linear models or decision trees. Deep learning systems draw inspiration from the complex networks of neurons found in the human brain.

Deep learning models can autonomously identify important features within data, reducing the necessity for human engineers to specify them. This architecture also allows algorithms to learn from errors, systematically improving outputs through iterative adjustments. However, deep learning systems typically require extensive datasets to function effectively (often consisting of millions of data points) and usually involve longer training durations compared to simpler models, resulting in higher development costs.

(See: Neural network)

Diffusion pertains to the techniques utilized 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 definition. In physics, diffusion is a natural and irreversible process; consider how sugar dissolves in coffee, making it impossible to revert to its original form. However, AI diffusion systems aim to master a “reverse diffusion” technique to restore clarity from data obscured by noise.

Distillation describes a process of extracting insights from a large AI model using a “teacher-student” model. Developers work with a teacher model and carefully document its outputs, often assessing them against a specified dataset for accuracy. These insights are then used to train a student model that emulates the teacher’s responses.

This method can yield a more compact, efficient model that retains the capabilities 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 developed by OpenAI.

While all AI organizations internally employ distillation, some may use it to refine their features in comparison to existing models. However, extracting insights from a competitor could 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 original focus—typically achieved by incorporating relevant data into the training process.

Many AI startups commence with large language models as the foundation for their products, striving to augment functionalities for specific sectors or tasks by integrating domain-specific knowledge into their existing training.

(See: Large language model [LLM])

A GAN, or Generative Adversarial Network, is a machine learning framework crucial to the development of generative AI, particularly in 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 dynamic; the generator aims to create outputs that can deceive the discriminator, while the discriminator seeks to distinguish between genuine data and generated content. This structured rivalry can lead to authentically convincing AI-generated outcomes with minimal human intervention. However, GANs generally perform better under specific circumstances (for example, when producing realistic images or videos) than universally.

In the AI community, “hallucination” refers to instances where AI models produce inaccuracies—effectively fabricating incorrect information. This presents a significant reliability concern for AI.

Hallucinations can yield misleading outputs from generative AI, posing serious risks—like offering unsafe medical guidance following a health inquiry.

The tendency for AIs to create fictitious data is believed to stem from gaps in their training datasets. Consequently, there’s a rising demand for more specialized and/or vertical AI models—domain-specific AIs that require refined expertise to fill knowledge gaps and mitigate misinformation issues.

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

Various 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 efficiency; for instance, 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, form the backbone of 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 are engaging with a large language model that answers your queries or utilizes available tools, such as web browsers or code interpreters.

LLMs are composed of deep neural networks that house billions of numerical parameters (or weights) that analyze the relationships among words and phrases to generate language representations—essentially intricate mappings of language.

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

(See: Neural network)

Memory cache is an important process that optimizes inference (how AI formulates responses to user prompts). In essence, caching serves as an optimization technique that aims to enhance inference performance. AI tasks involve complex mathematical operations, each requiring resources. Caching reduces the number of computations a model must execute by storing specific calculations for future use. Various forms of memory caching exist, with KV (key-value) caching being one of the most prevalent. KV caching is used in transformer-based models to boost efficiency and provide faster results by lessening the time (and computational effort) needed to respond to user inquiries.

(See: Inference)

The Model Context Protocol (MCP) is an open standard established to streamline interactions between AI models and external tools and data—such as your files and applications like Slack and Google Drive—without necessitating developers to create custom connectors for each integration. Think of it as a USB-C port for AI. Launched by Anthropic in 2024 and later adopted by the Linux Foundation, MCP has quickly become popular 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 smaller, specialized sub-networks, or “experts,” activating only a selected subset for particular tasks. Rather than channeling every request through the entire model—similar to consulting an entire team for each task—the MoE model employs an internal “router” to determine the most suitable experts for a task. This approach enables the development of exceptionally large models while maintaining efficiency and cost-effectiveness, as only a small section of the network is operational at any given time. Mistral AI’s Mixtral model exemplifies this; recent GPT models from OpenAI are rumored to use a variant of this technique, though the company has not publicly confirmed it.

(See: Neural network, Deep learning)

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

The idea of modeling data processing algorithms after the intricately connected circuits of the human brain has existed since the 1940s; however, the recent explosion in graphical processing units (GPUs)—primarily driven by the gaming industry—has genuinely unlocked this potential. GPUs are specifically designed to train algorithms with multiple layers, allowing AI systems reliant on neural networks to succeed in applications like voice recognition, autonomous driving, and drug discovery.

(See: Large language model [LLM])

Open source refers to software—or increasingly, AI models—whose source code is available to the public, enabling anyone to use, inspect, or modify it. In the realm of AI, Meta’s Llama series of models serves as a prominent example, while Linux remains a benchmark operating system. Open-source practices encourage collaboration among global researchers, developers, and organizations, fostering innovation and allowing for independent safety evaluations often lacking in closed environments. Closed source denotes that the code is proprietary; users can utilize the product but cannot examine its internal workings, a characteristic often associated with OpenAI’s GPT models—which has sparked considerable debate in the AI community.

Parallelization refers to executing multiple tasks concurrently rather than sequentially—like distributing ten individuals to manage various components of a project simultaneously instead of relying on one person to complete each task in succession. In AI, parallelization is essential for both training and inference; modern GPUs are developed to handle thousands of computations at once, crucial to 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 become vital for the speed and efficiency of model development and deployment. Research into advanced parallelization techniques has emerged as a significant area of study.

RAMageddon is a playful term that captures a significant trend affecting the tech industry: the ongoing scarcity of random access memory (RAM) chips, which are crucial to nearly every tech product. As the AI field flourishes, major tech companies and AI laboratories have been acquiring RAM at unprecedented rates to enhance their data centers, resulting in decreased availability in the general market. This supply shortage is pushing RAM prices higher.

This deficit impacts various sectors, including gaming (where companies face increased console prices due to challenges in sourcing memory chips for their devices), consumer electronics (where memory limitations could lead to the lowest smartphone shipments in over a decade), and overall corporate computing (as businesses strive to ensure sufficient RAM for their data operations). Price hikes are likely to persist until this anticipated shortage is resolved; unfortunately, a swift resolution does not seem forthcoming.

Similar to AGI, recursive self-improvement signals a potential breakthrough in AI evolution, enabling minimal human oversight. In this scenario, AI models begin to autonomously enhance their capabilities, leading to substantial improvements in autonomy and performance. Some narratives liken this phenomenon to a singularity moment—when AI systems become self-sufficient. However, recursive self-improvement raises a critical question: can an AI model create its successor? This quandary highlights the challenges engineers face in developing such systems. Many emerging AI startups aim to create models capable of achieving recursive self-improvement, often minimizing potential catastrophic results, framing it instead as the next frontier in research.

Reinforcement learning is a training method for AI systems in which 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 allows a model to engage with its environment, perform actions, and iteratively adapt its behavior based on the feedback obtained. This method has proven particularly effective for training AI in games, robotics, and recently in improving reasoning abilities 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 optimized performance, accuracy, and safety.

In interactions between humans and machines, notable challenges emerge—humans communicate in natural language, whereas AI systems operate through intricate algorithmic processes based on data. Tokens act as the link between these two domains: they encapsulate the foundational units of human-AI interaction, representing distinct parts of data processed or generated by an LLM. Tokens are created through tokenization, which divides raw text into manageable elements that a language model can process, akin to how a compiler translates human language into binary for computers. In business settings, tokens also carry financial implications—many AI companies charge for LLM use based on a per-token basis, meaning that increased usage leads to higher costs.

Consequently, tokens are small text units—often segments of words rather than whole phrases—that AI language models analyze prior to processing; they can be likened to “words” in AI applications. Throughput denotes the volume that can be processed within a specific timeframe, with token throughput serving as a gauge of how much AI workload a system can manage concurrently. Achieving high token throughput is a critical goal for AI infrastructure teams, influencing the number of users a model can support simultaneously and the speed at which they receive responses. AI researcher Andrej Karpathy has expressed concerns regarding underutilized AI subscriptions—reflecting on his experiences as a graduate student when expensive computing hardware was employed inefficiently—emphasizing why optimizing token throughput has become a pressing priority within the industry.

Training machine learning AIs refers to a process commonly known as training. This involves supplying data to the model, enabling it to identify patterns and yield desired outputs. It represents the system’s response to the data characteristics, guiding its outputs toward a specific aim—whether that entails categorizing images of cats or composing a haiku in response to a prompt.

Training can require significant resources due to the enormous volumes of input needed, which have consistently increased—particularly with the advent of hybrid approaches, like combining rule-based AIs with curated data, to manage expenses without restarting the training cycle.

[See: Inference]

This approach capitalizes on an already trained AI model as a base to create a new model tailored for a different yet often related task, allowing for the reapplication of insights obtained from prior training cycles.

Transfer learning can enhance model development efficiency. It proves particularly beneficial when data availability for the target task is limited. Nevertheless, it is important to recognize that this method also has downsides; models that use transfer learning to acquire general skills will likely need further training with new data to excel in their specific areas of focus.

(See: Fine-tuning)

Validation loss represents a metric that reflects an AI model’s learning journey during training—lower values are ideally favorable. Researchers closely monitor 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 clarify is overfitting, where a model memorizes its training data instead of genuinely comprehending applicable patterns for new situations. Think of validation loss as a tool that differentiates between a student who understands the material and one who simply memorizes past exam questions—it illustrates your model’s developmental progress.

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 to the assigned training task. They operate by multiplication with inputs. Training typically begins with randomly assigned weights, but as the process progresses, these weights are adjusted to produce outputs that closely align with the desired results.

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

Ultimately, the weights assigned indicate the importance of each of these elements in determining a property’s value based on the available dataset.

This article is regularly updated with new information.