6 Most Commonly Used AI Terms & What They Mean
Hello Folks,
I am on my journey to learn about Large Language Models (LLM’s) and thought to share most commonly used terms —
- Tokens
2. Context Window
3. Hallucinations
4. Retrieval-Augmented Generation
5. Grounding
6. Model Context Protocol
1. Tokens
Tokens are the fundamental units/smallest unit of text that AI models use to understand and process language.
Before an LLM can understand your input prompt, a process called tokenization converts the text/prompt into a sequence of tokens.
An example is OpenAI Tokenizer — Here the text input of 7 words (“Hello is World This is Agent Referencing”) has been converted into 8 Tokens.
We can see how each word has been converted into a Token.
In the next example below, we can see that a single word like “Referencing” has been assigned 2 Tokens.
As per this help article from OpenAI, here is how tokens are calculated
Helpful rules of thumb for English:
1 token ≈ 4 characters
1 token ≈ ¾ of a word
100 tokens ≈ 75 words
1–2 sentences ≈ 30 tokens
1 paragraph ≈ 100 tokens
~1,500 words ≈ 2,048 tokens
2. Context Window
The context window is like a language model’s working memory, determining how much of a conversation it can remember.
As per IBM, The context window (or “context length”) of a large language model (LLM) is the amount of text, in tokens, that the model can consider or “remember” at any one time. A larger context window enables an AI model to process longer inputs and incorporate a greater amount of information into each output.
For anyone looking to build their knolwedge, IBM Technologies Youtube Channel (especially Martin Keen) is an awesome source.
Get Ashish Verma’s stories in your inbox
Join Medium for free to get updates from this writer.
3. Hallucinations
An AI Hallucination is defined as the generation of incorrect, misleading, or fabricated information by a model that presents the output as fact.
The root causes often involve shortcomings in the training process, such as insufficient data, inherent biases in the source material, or faulty assumptions used by the model’s internal logic.
The practical risks of this phenomenon were recently highlighted when Deloitte’s Australian member firm had to issue a partial refund for a $290,000 report.
The document contained several alleged AI-generated errors, including references to non-existent academic research papers and a fabricated quote from a federal court judgment.
Another example of Hallucinations is from Andrej Karpathy’s (OpenAI Co-founder) Deep Dive into LLMs video, is where he asks the LLM for information on a non-existent person, and the model hallucinates a confident response.
This is a great 3-hour video for anyone looking to understand how LLMs are built and trained. Highly Recommended.
4. Retrieval-Augmented Generation (RAG)
RAG, or Retrieval-Augmented Generation is a specific architectural method for generating information by fetching it from a private knowledge base (such as company database, a set of internal documents, or a PDF library).
Why RAG — At the time of writing this piece, LLMs are limited to their pre-trained data.
This leads to outdated and potentially inaccurate responses. RAG overcomes this by providing up-to-date information to LLMs.
For example, consider a smart chatbot that can answer human resource questions for an organization. If an employee searches, “How much annual leave do I have?” the system will retrieve annual leave policy documents alongside the individual employee’s past leave record.
These specific documents will be returned because they are highly-relevant to what the employee has input. The relevancy was calculated and established using mathematical vector calculations and representations.
5. Grounding
Grounding is a general objective or process of tying the LLM’s response to verifiable, factual information.
If you provide models with access to specific data sources, then grounding tethers their output to these data and reduces the chances of inventing content. This is particularly important in situations where accuracy and reliability are significant.
In the example below, we can view search queries which GPT is using to go out in the web and extract information in order to return to the users. Not every prompt needs a web search due to the way LLMs are built.
In the example below, you will see the prompt on the left-hand side while Chrome console displays queries.
6. Model Context Protocol (MCP)
Introduced by Anthropic in November 2024, MCP provides a secure and standardized “language” for LLMs to communicate with external data, applications, and services.
It acts as a bridge, allowing AI to move beyond static knowledge and become a dynamic agent that can retrieve current information and take action, making it more accurate, useful, and automated.
In short, MCP acts like a “USB-C port for AI applications,” providing a universal, standardized, and secure way for an LLM to find, connect to, and execute functions across diverse external systems and tools.
Looking forward to building this knowledge base with you.
