Demystifying AI: A Simple Guide to Common Terms, from LLMs to Hallucinations

The world of Artificial Intelligence (AI) can often feel like navigating a labyrinth, filled with complex jargon and technical terms. To help demystify this rapidly evolving field, we’ve compiled a glossary of essential AI terms, designed to provide clarity and understanding for everyone.

This guide, regularly updated with the latest advancements and safety considerations, will serve as your go-to resource for navigating the AI landscape.

AGI (Artificial General Intelligence)

Artificial General Intelligence (AGI) remains a somewhat ambiguous concept. It generally describes AI systems possessing capabilities exceeding those of an average human across a broad spectrum of tasks. Sam Altman, CEO of OpenAI, has described AGI as the “equivalent of a median human that you could hire as a co-worker.” OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind envisions AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Despite differing perspectives, experts at the forefront of AI research acknowledge the evolving nature of AGI. [1]

AI Agent

An AI agent is a tool utilizing AI technologies to execute a series of tasks on your behalf. These tasks extend beyond the capabilities of basic AI chatbots, potentially including managing expenses, making reservations, or even coding. While the space is still developing, the core concept involves an autonomous system leveraging multiple AI systems to complete multistep processes. [2]

Chain of Thought

Chain-of-thought reasoning, in the context of AI, involves breaking down complex problems into smaller, more manageable steps to enhance the accuracy of the final result. This approach is particularly useful in logic and coding contexts, where intermediary steps are crucial for arriving at the correct solution. Reasoning models are often developed from large language models and optimized for chain-of-thought thinking through reinforcement learning.

Deep Learning

Deep learning is a subset of machine learning where AI algorithms are designed with multi-layered, artificial neural network (ANN) structures. This allows them to discern more complex correlations in data compared to simpler machine learning systems. Inspired by the interconnected pathways of neurons in the human brain, deep learning models can identify important characteristics in data without explicit human instruction. However, these systems require vast amounts of data (millions of data points or more) and significant computational resources, leading to higher development costs.

Diffusion

Diffusion is a core technology behind many AI models that generate art, music, and text. Diffusion systems operate by gradually adding noise to data, such as photos or songs, until its original structure is unrecognisable. The AI then learns a “reverse diffusion” process to restore the data from the noise, effectively gaining the ability to recreate the original data. [3]

Distillation

Distillation is a technique used to create a smaller, more efficient AI model based on a larger, more complex model. This process involves training a ‘student’ model to mimic the behavior of a ‘teacher’ model. While widely used internally by AI companies, using distillation to copy a competitor’s model may violate terms of service.

Fine-tuning

Fine-tuning involves further training an AI model to optimize its performance for a specific task or area. This is typically achieved by feeding the model new, specialized data. Many AI startups leverage large language models and then fine-tune them with domain-specific knowledge to enhance their utility for target sectors or tasks.

GAN (Generative Adversarial Network)

A GAN, or Generative Adversarial Network, is a machine learning framework used in generative AI to produce realistic data, including deepfakes. GANs consist of two neural networks: a generator that creates data and a discriminator that evaluates it. The two networks compete against each other, improving the realism of the generated data without human intervention. However, GANs are best suited for specific applications, such as generating realistic photos or videos, rather than general-purpose AI.

Hallucination

Hallucination is the term used when AI models generate incorrect or fabricated information. This poses a significant challenge to AI quality, as it can lead to misleading or even dangerous outputs. The problem often arises from gaps in training data, making it difficult for general-purpose AI to resolve all possible queries accurately. Hallucinations are driving the push towards more specialized AI models with narrower expertise to reduce disinformation risks.

Inference

Inference is the process of running an AI model to make predictions or draw conclusions from data. It requires prior training to enable the model to extrapolate from learned patterns. Different types of hardware can perform inference, but the speed and efficiency vary depending on the model size and hardware capabilities.

Large Language Model (LLM)

Large language models (LLMs) are used by popular AI assistants like ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, and Mistral’s Le Chat. LLMs are deep neural networks with billions of parameters that learn relationships between words and phrases to create a representation of language. These models are trained on vast amounts of text data, enabling them to generate likely patterns in response to prompts.

Neural Network

A neural network is the multi-layered algorithmic structure that underpins deep learning and generative AI. Inspired by the human brain, neural networks consist of interconnected pathways for data processing. The rise of graphical processing hardware (GPUs) has significantly advanced neural network capabilities, enabling AI systems to achieve better performance in areas like voice recognition and autonomous navigation.

Training

Training is the process of feeding data into a machine learning AI model so it can learn from patterns and generate useful outputs. This process shapes the AI model by enabling it to adapt outputs towards a specific goal. While not all AI requires training, self-learning systems typically offer more flexibility and capability. Training can be expensive due to the large volumes of data required.

Transfer Learning

Transfer learning is a technique where a previously trained AI model is used as the starting point for developing a new model for a related task. This approach can save time and resources by reapplying knowledge gained in previous training cycles. However, models relying on transfer learning may require additional training on specific data to perform well in their target domain.

Weights

Weights are numerical parameters that determine the importance given to different features in the data used for training an AI system. They define what is most salient in a dataset and influence the AI model’s output. Model training typically begins with randomly assigned weights, which are adjusted as the model learns to match the target output more closely.