AI for Altruism

AI for AltruismAI for AltruismAI for Altruism
Home
Grants
Resources

AI for Altruism

AI for AltruismAI for AltruismAI for Altruism
Home
Grants
Resources
More
  • Home
  • Grants
  • Resources
  • Home
  • Grants
  • Resources

AI Lexicon

Alignment

The broader goal of ensuring AI systems behave in ways that are beneficial and aligned with human values and intentions. This includes technical approaches like RLHF, but also encompasses broader considerations about safety, ethics, and the responsible development of AI systems.

Distillation

A technique where a smaller model (student) is trained to mimic the behavior of a larger model (teacher). The student learns to approximate the teacher's capabilities while being more efficient in terms of size and computational requirements. This makes deployment more practical and cost-effective.

Fine-Tuning

The process of further training a pre-trained LLM on a specific dataset for a particular task or domain. This allows the model to adapt its knowledge and capabilities to specialized use cases while maintaining its general capabilities. It's like giving additional specialized training to an already educated model.

GPT (Generative Pre-trained Transformer)

A specific architecture of AI model that's trained to predict the next token (word/character) in a sequence. It's "generative" because it can create new content, "pre-trained" because it's first trained on a broad dataset before any specific fine-tuning, and uses the "transformer" architecture as its foundation.

Hallucination

Refers to when language models generate content that is false, inaccurate, or completely fabricated, despite appearing confident and fluent. This can happen because LLMs are trained to predict statistically likely patterns in text rather than to represent verified truth, and they may fill in gaps in their knowledge with plausible-sounding but incorrect information.

LLM (Large Language Model)

An AI model trained on vast amounts of text data to understand and generate human language. These models process and generate text by predicting the most probable next words in a sequence, based on their training. Examples include GPT-4, Claude, and LLaMA. They can perform tasks like writing, analysis, coding, and conversation.

LMM (Large Multimodal Model)

A more recent evolution that can understand and work with multiple types of input (modes) - not just text, but also images, audio, and sometimes video. These models can process different types of media simultaneously and generate responses across modes. For example, they can describe images, generate images from text descriptions, or understand the context of a conversation that includes both text and images.

RAG (Retrieval Augmented Generation)

A technique that combines an LLM with an external knowledge base. When given a query, the system retrieves relevant information from its database and uses it to generate more accurate, up-to-date responses. This helps address LLMs' hallucination issues and knowledge cutoff limitations.

RLHF (Reinforcement Learning from Human Feedback)

A training method where human feedback is used to refine an LLM's outputs. The model is rewarded for generating responses that align with human preferences and penalized for undesirable outputs. This helps the model learn to be more helpful, truthful, and safe.

Transformer

The fundamental neural network architecture that revolutionized natural language processing. Its key innovation is the "attention mechanism," which allows the model to weigh the importance of different parts of the input when generating each part of the output. It can process all elements of a sequence in parallel, making it more efficient and better at capturing long-range dependencies in data.

AI Chat - Short Courses

Objective: Understand what instruction-tuned language models are and their basic capabilities.

Background:

Instruction-tuned large language models (LLMs) are advanced AI systems trained to follow user instructions. ChatGPT, Claude, and Gemini are common examples.  They are versatile and can perform tasks such as answering questions, writing essays, solving math problems, or summarizing text.

These models are designed to:

  1. Understand Instructions: Respond clearly and accurately to user-provided prompts.
  2. Adapt: Generate responses based on the tone, style, or specifics of your request.
  3. Perform Broad Tasks: Handle diverse domains like writing, coding, education, and more.

Practical Examples:

Basic Prompting:

  1. Task: Ask for information.

  • Prompt: “Explain photosynthesis in simple terms.”
  • Model Response: Photosynthesis is the process by which plants use sunlight, carbon dioxide, and water to create food and release oxygen.

  1. Task: Ask for step-by-step instructions.

  • Prompt: “How do I bake a chocolate cake?”
  • Model Response: [Provides a step-by-step recipe.]

Activity: Try writing 3-5 simple prompts for various tasks (e.g., asking for definitions, instructions, or lists).

Objective: Learn how to write clear and structured prompts to get better responses.

Background:

The quality of the response depends on the clarity and specificity of your prompt. Ambiguous prompts may yield vague answers, while detailed prompts provide precise results.

Tips for Effective Prompts:

  1. Be Specific: Clearly state what you want.
  2. Provide Context: Include background or examples if needed.
  3. Ask Directly: Avoid open-ended or overly vague questions.

Practical Examples:

  1. Specific vs. Vague Prompt:

  • Vague: “Tell me about AI.”
  • Specific: “What are the main applications of AI in education?”

  1. Adding Context:

  • Without Context: “Explain gravity.”
  • With Context: “Explain gravity as if teaching a 10-year-old.”

  1. Requesting Output Style:

  • Prompt: “Write a poem about winter in a cheerful tone.”
  • Model Response: [Generates a cheerful winter poem.]

Activity: Rephrase three vague prompts into specific, detailed ones and compare the responses.

 

Objective: Understand the range of tasks LLMs can perform and how to tailor prompts for each type.

Background:

Instruction-tuned LLMs can assist with:

  • Creative Writing: Stories, poems, scripts.
  • Educational Content: Summaries, explanations, or lesson plans.
  • Problem Solving: Math, logic, coding.
  • Language Translation: Translating text between languages.

Practical Examples:

  1. Summarization:

  • Prompt: “Summarize the main points of a news article about climate change.”
  • Model Response: [Generates a concise summary.]

  1. Creative Writing:

  • Prompt: “Write a short story about a robot learning to paint.”
  • Model Response: [Creates a narrative.]

  1. Problem Solving:

  • Prompt: “Solve this math problem: What is 25% of 160?”
  • Model Response: 40.

Activity: Try one prompt from each task category (creative, educational, problem-solving, translation) and analyze the responses.

Objective: Learn how to refine prompts and iterate based on initial responses.

Background:

Sometimes, the model’s response might not fully meet your expectations. Refining the prompt or asking follow-up questions can lead to better outcomes.

Techniques for Refinement:

  1. Provide Examples: Include samples of desired outputs.
  2. Specify Format: Request lists, paragraphs, or structured answers.
  3. Follow Up: Adjust the prompt or ask clarifying questions.

Practical Examples:

  1. Refinement Through Examples:

  • Initial Prompt: “Describe a character.”
  • Revised Prompt: “Describe a character in detail, including their appearance, personality, and background.”

  1. Requesting Format:

  • Prompt: “List five ways to improve productivity.”
  • Model Response: [Generates a numbered list.]

  1. Follow-Up Questions:

  • Initial Prompt: “What is the best way to learn a new language?”
  • Follow-Up: “Can you list apps or tools to help with this?”

Activity: Start with a basic prompt, refine it based on the response, and try adding follow-up prompts to dig deeper into the topic.

Objective: Use advanced strategies to create precise, multi-step prompts for more sophisticated outputs.

Background:

As tasks become more complex, prompt structuring becomes critical. Advanced prompts often involve:

  • Chaining Instructions: Requesting multiple steps or elements in one prompt.
  • Setting Context First: Framing the task with a preamble or background.
  • Using Variables or Parameters: Giving conditions or constraints.

Practical Examples:

  1. Chaining Instructions:

  • Prompt: “Explain Newton’s laws of motion briefly, and then give an example of each.”
  • Model Response: [Provides a summary and examples for all three laws.]

  1. Setting Context:

  • Prompt: “Imagine you are an environmental scientist. Explain the effects of deforestation in a way that high school students can understand.”
  • Model Response: [Tailors the response to the audience and context.]

  1. Using Constraints:

  • Prompt: “List three ideas for a science fair project that require a budget under $20.”
  • Model Response: [Generates budget-friendly ideas.]

Activity: Write 2-3 chained or constrained prompts and observe how the responses align with your instructions.

Objective: Learn how to guide the model through iterative interactions to improve responses.

Background:

Sometimes the initial response might need adjustments. Iterative feedback helps you refine outputs step-by-step by:

  • Asking Follow-Up Questions: Clarify or expand on specific parts.
  • Requesting Revisions: Direct changes to the format or content.
  • Exploring Alternatives: Ask for different versions of the same task.

Practical Examples:

  1. Follow-Up Question:

  • Initial Prompt: “Describe the benefits of remote work.”
  • Follow-Up: “Can you add disadvantages for balance?”

  1. Requesting Revisions:

  • Initial Prompt: “Write a poem about autumn.”
  • Follow-Up: “Can you rewrite it in a humorous tone?”

  1. Exploring Alternatives:

  • Prompt: “Suggest a title for a mystery novel.”
  • Follow-Up: “Can you provide three more options?”

Activity: Start with a prompt, refine it through two iterations, and compare the initial and final responses.

Objective: Understand how to tailor prompts for different types of tasks more effectively.

Background:

Different tasks require specific prompt strategies to optimize results. These tasks include:

  • Data Extraction: Pulling specific information from text.
  • Comparative Analysis: Asking for comparisons between options.
  • Role-Playing: Instructing the model to adopt a persona for tailored responses.

Practical Examples:

  1. Data Extraction:

  • Prompt: “Extract the key dates and events from this text: [insert text].”
  • Model Response: [Lists the dates and corresponding events.]

  1. Comparative Analysis:

  • Prompt: “Compare the pros and cons of solar energy and wind energy.”
  • Model Response: [Provides a clear side-by-side comparison.]

  1. Role-Playing:

  • Prompt: “You are a career coach. How would you advise someone transitioning from teaching to a corporate role?”
  • Model Response: [Generates personalized advice in the voice of a coach.]

Activity: Try one prompt for each task type and analyze how effectively the model performs.

Objective: Learn how to use the model as a partner for brainstorming and iterative idea development.

Background:

LLMs can be powerful tools for collaboration, acting as sounding boards or co-creators. This involves:

  • Brainstorming Ideas: Generate a wide range of possibilities.
  • Refining Solutions: Build on initial ideas to create polished results.
  • Proposing Alternatives: Evaluate and refine multiple approaches.

Practical Examples:

  1. Brainstorming:

  • Prompt: “Give me 10 unique ideas for a community service project.”
  • Model Response: [Generates a diverse set of ideas.]

  1. Refining Solutions:

  • Initial Prompt: “Suggest an idea for a small business.”
  • Follow-Up: “What are the potential challenges of this idea, and how can they be addressed?”

  1. Proposing Alternatives:

  • Prompt: “Suggest different ways to teach fractions to elementary school students.”
  • Follow-Up: “Can you suggest a more interactive approach?”

Activity: Use the model to brainstorm a topic of your choice, refine the ideas, and propose alternative solutions.

Objective: Leverage the model’s contextual memory to conduct complex, multi-turn dialogues.

Background:

Instruction-tuned LLMs can handle extended conversations where previous inputs influence subsequent responses. This enables:

  • Building Context: Carrying information from one prompt to the next.
  • Dynamic Refinement: Adjusting the direction of a conversation based on evolving needs.
  • Simulating Real-World Interactions: Role-playing or simulating scenarios across multiple turns.

Practical Examples:

  1. Building Context:

  • Prompt 1: “Explain the lifecycle of a butterfly.”
  • Prompt 2: “Now, compare this to the lifecycle of a frog.”
  • Model Response: [Draws on the butterfly explanation to create a comparison.]

  1. Dynamic Refinement:

  • Prompt 1: “Outline a marketing strategy for a tech startup.”
  • Prompt 2: “Can you expand the strategy to include social media campaigns?”
  • Model Response: [Builds on the initial strategy.]

  1. Simulating Real-World Interactions:

  • Prompt: “Pretend to be a job interviewer. Ask me three questions for a software engineering role.”
  • Model Response: [Provides realistic interview questions.]

Activity: Create a multi-turn conversation, starting with a general topic and refining it step-by-step across three or more prompts.

Objective: Instruct the model to adopt specific roles, perspectives, or stylistic approaches.

Background:

Models can mimic styles or personas when explicitly directed. This allows for:

  • Role-Based Interactions: Interacting as a professional, teacher, or fictional character.
  • Stylistic Adaptation: Adopting tones like formal, casual, persuasive, or poetic.
  • Specialized Perspectives: Writing from a historical, scientific, or creative viewpoint.

Practical Examples:

  1. Role-Based Interaction:

  • Prompt: “You are a financial advisor. What investment strategies would you recommend to someone with a low risk tolerance?”
  • Model Response: [Generates a tailored financial plan.]

  1. Stylistic Adaptation:

  • Prompt: “Write a review of this product in a humorous tone.”
  • Model Response: [Creates a lighthearted, funny review.]

  1. Specialized Perspectives:

  • Prompt: “Write a letter to a friend as if you were Albert Einstein, explaining relativity.”
  • Model Response: [Emulates Einstein’s perspective and voice.]

Activity: Try role-playing or stylistic prompts, asking for responses in different tones or from specialized viewpoints.

Objective: Use task chaining to solve multi-step problems and automate workflows.

Background:

Task chaining involves breaking down a complex goal into smaller, sequential tasks that the model can handle step-by-step. This is useful for:

  • Research and Summarization: Collecting, organizing, and synthesizing information.
  • Creative Processes: Developing ideas iteratively, from brainstorming to execution.
  • Complex Problem-Solving: Tackling challenges that require logical sequencing.

Practical Examples:

  1. Research Workflow:

  • Prompt 1: “Find the main arguments for and against renewable energy adoption.”
  • Prompt 2: “Summarize these arguments in bullet points.”
  • Prompt 3: “Draft a balanced conclusion based on the summary.”

  1. Creative Process:

  • Prompt 1: “Generate 10 potential names for a fantasy novel.”
  • Prompt 2: “Select the best three and suggest a plot summary for each.”

  1. Problem-Solving:

  • Prompt 1: “Plan a week-long itinerary for a family visiting Paris.”
  • Prompt 2: “Add budget considerations for meals and transportation.”

Activity: Develop a task chain for a topic of your choice, starting with research, moving to synthesis, and ending with actionable insights.

Objective: Use detailed customization techniques to achieve highly specific outputs.

Background:

Advanced users can guide the model to:

  • Format Outputs Precisely: Request tables, outlines, or formatted text.
  • Incorporate Constraints: Limit outputs by word count, style, or content focus.
  • Generate High-Fidelity Outputs: Ensure accuracy and relevance through iterative refinement.

Practical Examples:

  1. Precise Formatting:

  • Prompt: “Create a pros and cons table for electric vehicles.”
  • Model Response: [Generates a neatly formatted table.]

  1. Incorporating Constraints:

  • Prompt: “Write a 150-word summary of the French Revolution, focusing only on its causes.”
  • Model Response: [Produces a concise, focused summary.]

  1. High-Fidelity Outputs:

  • Prompt: “List the steps to set up a small business, ensuring compliance with U.S. tax laws.”
  • Follow-Up: “Can you elaborate on the tax compliance steps?”

Activity: Use constraints or formatting requirements in your prompts and compare the results with less specific versions.

AI Chat - Practical Examples

Use Case: Simplify complex concepts for learners of all ages.

Scenario: Create a simplified explanation of climate change for a 10-year-old.

  1. Step 1: Start with a clear, basic prompt.

  • Prompt: "Explain climate change in simple terms for a 10-year-old."

  1. Expected Response:

  • A concise, age-appropriate explanation that uses simple words and relatable examples.

  1. Step 2: Add follow-up questions to refine understanding.

  • Prompt: "Can you add an example of how climate change affects animals?"

Result: A comprehensive yet easy-to-understand explanation with concrete examples.

Use Case: Provide general advice for managing stress, while acknowledging the importance of professional help.

Scenario: Offer stress management techniques for someone feeling overwhelmed.

  1. Step 1: Craft a thoughtful and supportive prompt.

  • Prompt: "Suggest some healthy and practical stress management techniques that anyone can use."

  1. Expected Response:

  • Suggestions like deep breathing, journaling, taking walks, and practicing mindfulness.

  1. Step 2: Ask for a tailored response.

  • Prompt: "Focus on techniques that work for individuals with limited free time."

  1. Step 3: Ensure responsible advice.

  • Prompt: "Please include a disclaimer emphasizing the importance of seeking professional help if stress becomes overwhelming."

Result: A thoughtful, actionable, and socially responsible resource.

 

Use Case: Encourage informed participation in local elections.

Scenario: Provide a guide for researching candidates and policies.

  1. Step 1: Start with a broad prompt.

  • Prompt: "How can someone research candidates and policies before voting in an election?"

  1. Expected Response:

  • General steps like visiting official websites, reading news articles, and checking candidates' track records.

  1. Step 2: Request specific examples.

  • Prompt: "List trusted online resources for researching U.S. elections."

  1. Step 3: Tailor the advice to accessibility.

  • Prompt: "How can someone with limited internet access stay informed about candidates?"

Result: A comprehensive guide that supports civic participation for diverse audiences.

Use Case: Create an emergency preparedness checklist for families in flood-prone areas.

Scenario: Develop a step-by-step plan for families to prepare for flooding.

  1. Step 1: Set the scope of the response.

  • Prompt: "Create a detailed emergency preparedness checklist for families living in areas prone to flooding."

  1. Expected Response:

  • A list covering essentials like emergency kits, evacuation routes, and communication plans.

  1. Step 2: Add constraints for inclusivity.

  • Prompt: "Include tips for families with young children or elderly members."

  1. Step 3: Address post-disaster needs.

  • Prompt: "What steps should families take immediately after a flood to ensure safety and recover?"

Result: A well-rounded, actionable plan tailored to diverse family needs.

Use Case: Create a persuasive campaign for reducing plastic waste in local communities.

Scenario: Generate creative ideas for an awareness campaign about reducing single-use plastics.

  1. Step 1: Ask for a brainstorming list.

  • Prompt: "Suggest five creative ideas for a campaign to reduce single-use plastics in a community."

  1. Expected Response:

  • Ideas such as reusable bag giveaways, educational workshops, and art installations made from recycled materials.

  1. Step 2: Request detailed execution plans.

  • Prompt: "Provide a step-by-step guide for organizing a reusable bag giveaway at a local farmers' market."

  1. Step 3: Make it data-driven.

  • Prompt: "What statistics about plastic waste could be included in this campaign to make it more impactful?"

Result: A data-informed, actionable campaign plan that inspires community involvement.

Use Case: Create educational content about preventing the spread of infectious diseases.

Scenario: Develop an infographic script for preventing the spread of seasonal flu.

  1. Step 1: Outline key points.

  • Prompt: "List the most effective ways to prevent the spread of seasonal flu."

  1. Expected Response:

  • Tips such as washing hands, wearing masks, and getting vaccinated.

  1. Step 2: Ask for an infographic format.

  • Prompt: "Reformat this information as an infographic script with headings, bullet points, and short sentences."

  1. Step 3: Tailor for different audiences.

  • Prompt: "Simplify this for an audience of middle school students."

  1. Step 4: Add multilingual support.

  • Prompt: "Translate this infographic script into Spanish."

Result: A multilingual, audience-specific script for educational purposes.

 AI Beyond Work

Who says AI is just for serious tasks? Let’s explore some creative, fun, and even quirky ways you can use AI:

AI Art Creation

Example:
"Generate a surrealist-style poster promoting our AI workshop!"

Collaborative Storytelling

Example:
"Help me co-write a short story where a group of students use AI to solve a mystery at school."

Music Mashups

Example:
"Suggest three AI music tools and help me create a playlist of tracks for our community event."

Brainstorm Team-Building Activities

Example:
"Design an AI-themed scavenger hunt for our nonprofit staff retreat."

AI isn’t just powerful—it’s playful. Dive in and see what unique creations you can bring to life!

Visit our software repository

https://github.com/ai4altruism

This page includes content generated with AI, reviewed for alignment by our team.

AI for Altruism (A4A) is a 501(c)(3) nonprofit organization.  

Copyright © 2025 AI for Altruism, Inc. - All Rights Reserved.

Please donate today.

  • Privacy Policy
  • Resources
  • Donation Policy

This website uses cookies.

We use cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.

Accept