I wrote some quick definitions for AI jargon because I needed a clear understanding of these phrases myself. The definitions capture the most commonly agreed explanations.

ARTIFICIAL INTELLIGENCE aka AI - You all should know this one but I’m defining it here as I want to clearly differentiate it from AGI below. AI refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. This includes problem-solving, recognizing patterns, understanding language, and performing tasks that typically require human intelligence.

ARTIFICIAL GENERAL INTELLIGENCE aka AGI - What we should all be at least a little wary of - AGI, also known as strong AI or full AI, refers to a type of AI that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a level equal to or surpassing that of a human being. AGI can perform any intellectual task that a human can, demonstrating generalized cognitive abilities.

GENERATIVE ARTIFICIAL INTELLIGENCE aka GenAI refers to AI systems that can generate new content, such as text, images, music, and other media, based on patterns and data they have been trained on. Examples include language models like GPT-3 and image generation models like MidJourney.

You will need to know what a Token is to put together your business plan. In Step 6, I go over the costs of each model for Token usage.

TOKEN - Tokens can be thought of as pieces of words. Before the API processes the request, the input is broken down into tokens. These tokens are not cut up exactly where the words start or end - tokens can include trailing spaces and even sub-words. Here are some helpful rules of thumb for understanding tokens in terms of lengths:

· 1 token ~= 4 chars in English

· 1 token ~= ¾ words

· 100 tokens ~= 75 words

Or

· 1-2 sentence ~= 30 tokens

· 1 paragraph ~= 100 tokens

· 1,500 words ~= 2048 tokens 

TOKEN LIMITS - Depending on the model used, requests can use up to 128,000 tokens (in GPT-4) shared between prompt and completion. Some models, like GPT-4 Turbo, have different limits on input and output tokens.

OUTPUT TOKEN LIMIT - Limit refers to the maximum number of tokens that an AI model can generate in a single session. In simpler terms, it's the total length of response you can get from an AI model, which contributes to the overall Token Limit.

CONTEXT WINDOW - denotes the total number of tokens permissible in a session. This includes tokens from a user's interaction with a chatbot in a conversational exchange or the processing of code, files, images, videos, etc. It accounts for both the user's input and the model's output.

Note: These * marked definitions were taken from OpenAI Help Community directly without edits.

OpenAI aka Chat GPT

Google Gemini

Anthropic Claude

Meta Llama

Go to Step 6 to see the Comparison Table of each one’s strength and weakness

Large Language Models (LLMs) are powerful tools, but their success hinges on how effectively you guide them. Here's a comprehensive guide to crafting prompts that unlock their full potential:

1. n-shot Prompts + In-Context Learning:

Concept: Train the model by providing it with multiple examples of the task you want it to perform. This helps it understand your expectations and generate more aligned responses.

Tips:

• Quantity matters: Aim for at least 5 examples for optimal generalization.

• Diversity is key: Include examples with varying styles, complexities, and lengths to enhance the model's adaptability.

• Real-world context: Use examples relevant to your specific task and domain.

Example:

**Task:** Translate English phrases into Spanish.

**Examples:**

* English: "Hello, how are you?" | Spanish: "Hola, ¿cómo estás?"

* English: "I am doing well, thank you." | Spanish: "Estoy bien, gracias."

* English: "What is your name?" | Spanish: "¿Cuál es tu nombre?"

* English: "My name is [Name]." | Spanish: "Mi nombre es [Name]."

* English: "Nice to meet you." | Spanish: "Encantado de conocerte."

**Prompt:** Translate the following English phrase into Spanish: "It's a pleasure to meet you."

2. Chain-of-Thought (CoT) Prompting:

Concept: Encourage the model to articulate its reasoning process before providing the final answer. This helps in understanding its thought process and can lead to more robust and insightful responses.

Tips:

• Explicitly ask for reasoning: Use phrases like "Explain your thought process," "Show your work," or "Think step by step."

• Guide the reasoning: Provide prompts that guide the model through specific steps or logical reasoning patterns.

Example:

**Task:** Determine the next number in the sequence: 2, 4, 8, 16...

**Prompt:**

* First, identify the pattern in the sequence.

* Then, apply the pattern to determine the next number.

* Finally, write the next number in the sequence.

3. Retrieval-Augmented Generation (RAG):

Concept: Enhance the model's accuracy and minimize errors by providing it with relevant text snippets to refer to during response generation.

Tips:

• Prioritize relevant information: Guide the model to focus on the provided snippets by using phrases like "Refer to the following text," or "Use these snippets to support your answer."

• Structure the snippets: Organize the snippets into a logical format, like a list or a table, for easier access.

Example:

**Task:** Answer a question about a specific topic using information from a provided document.

**Prompt:**

* Read the following document: [Insert document here].

* Answer the question: [Insert question here].

* Use information from the document to support your answer.

4. Structured Input and Output:

Concept: Improve communication and ensure consistent responses by clearly defining the format for both inputs and outputs.

Tips:

• JSON or Markdown: Use these formats to structure complex information and represent relationships between data points.

• Define schema: Specify the expected structure of the output, including fields, data types, and relationships.

Example:

**Task:** Summarize product reviews.

**Input format:** JSON array of product reviews, each containing fields like "review_text" and "rating".

**Output format:** JSON object with fields like "overall_sentiment" (positive, negative, neutral) and "key_themes" (list of common themes).

5. Specific Instructions:

Concept: Provide clear and detailed instructions to guide the model towards the desired outcome.

Tips:

• Break down complex tasks: Divide complex tasks into smaller, manageable subtasks for clearer instructions.

• Include steps or rules: Structure instructions in a step-by-step format or provide specific rules to follow.

Example:

**Task:** Write a formal letter of complaint.

**Prompt:**

* Start with a formal salutation and introduce yourself.

* Clearly state the purpose of the letter and the specific complaint.

* Provide detailed evidence and supporting information.

* Conclude with a request for resolution and a polite closing.

Use code with caution.

6. Small, Focused Prompts:

Concept: Break down complex tasks into smaller, focused prompts to improve clarity and accuracy.

Tips:

Address one subtask per prompt: Avoid overloading the model with multiple tasks in a single prompt.

Build on previous responses: Use the output of a previous prompt as input for the next one.

Example:

**Task:** Summarize a research paper.

**Prompt 1:** Extract the key points from the introduction of the research paper.

**Prompt 2:** Summarize the methodology section of the research paper.

**Prompt 3:** Summarize the results and conclusion of the research paper.

**Prompt 4:** Combine the summaries from prompts 1-3 to create a concise summary of the entire research paper.

By applying these prompt engineering techniques, you can unlock the true potential of LLMs and achieve better results in your applications. Remember, the key is to think like an engineer – design prompts that guide the model towards the desired outcome with clarity and precision.