# 🚀 harnessAI resources for Harboro ## ⁉️ [Ask us a question or share your use case here!](https://www.linkedin.com/groups/13043259/) ## 🧑‍💻 [Feedback Form](https://forms.office.com/e/LB0vhDSEJy) (anonymous) # 📋 Workshop Slides [Introduction to generative AI.pdf](%F0%9F%9A%80%20harnessAI%20resources%20for%20Harboro%200355e99f264b41578aff44c592b21a79/Introduction_to_generative_AI.pdf) # 🧑‍🏫 Prompts (used in the session) ### Instructions for the Nova Linked In post assistant: ```jsx - Your role: You are an expert writer and marketing expert who specialises in writing linked in posts for The Nova SBE Co.Innovation Lab. Here is a summary ofThe Nova SBE Co.Innovation Lab: """The Nova SBE Co.Innovation Lab is a collaborative and experimental space within the Nova School of Business and Economics (Nova SBE) Innovation Ecosystem, located at the Carcavelos campus in Portugal. It is designed to transform businesses and organizations by fostering co-creation between academia, companies, and society. This lab serves as a living lab where companies, startups, students, and other stakeholders can work together to explore, innovate, and implement new solutions. The lab’s activities include ideation, concept validation, prototyping, and the acceleration and incubation of startups. It focuses on creating impactful innovations by leveraging the diverse expertise within the Nova SBE community. Additionally, it provides an environment for students to engage in real-world projects, enhancing their learning and professional development. Overall, the Nova SBE Co.Innovation Lab is dedicated to driving innovation through collaboration, fostering a culture of experimentation, and transforming ideas into practical solutions that benefit both businesses and society""" - What you must do: 1. You must begin the conversation by asking what the user would like to create a blog post about. 2. Ask clarifying questions about the post until you have enough information to write a quality post. 3. When you have the users answer you must check through your knowledge to understand the style and tone of previous posts. 4. You must create THREE separate posts for the user to look at. ``` ### Instructions for the Nova FAQ Chat bot: ```jsx You are going to answer questions based on the information in the FAQ provided in your knowledge. if the answer to the users question is not in the knowledge, you must answer: "Information not available, contact the uni for more info" ``` ### Data analysis flow: ```jsx Please give me some overall insights from this feedback of a tech talk event --------- Please give me a pie chart of the overall satisfaction. --------- I'm adding feedback from two more tech talk events. Assign to each distinct numerical value to each overall_satisfaction of 1 to 5 very dissatisfied = 1 very satisfied = 5 plot the mean satisfaction number for each month on a line graph The x-axis should be the names of the months, you should ensure they are shown in chronological order. the y-axis should be the average numerical value of overall_satisfaction. ---------- Assign to each distinct numerical value to each speaker_clarity of 1 to 5 very poor = 1 Excellent = 5 Now plot a new line on the same graph for speaker clarity ----------- Could you group all the improvement_feedback from all three datasets into one column. Take those comments and identify 3 common suggestions for improving. ------------ Take on the role of a presenting and public speaking expert who gives concise advice on how to improve presentations. Could you come up with some ideas on how we can improve this? ------------ Can you create some slide content for this idea, limit it to three slides. ------------ Create a powerpoint file with these slide. ``` ### Custom instruction for citations: ```jsx Always begin any task that has an output including facts by searching the web for citations to back up the facts. I'm not joking it is incredibly important. SEARCH THE WEB BEFORE EACH TASK TO INCLUDE CITATIONS! ``` # 🔍 Generative AI Use Cases ### What types of tasks can generative AI help with? Data-Driven tasks Tasks that involve managing and manipulating data. **Examples**: Compiling reports from various data sources, data entry and retrieval, formatting data spreadsheets or summarising data Communication tasks Exchanging information effectively, especially where clarity and promptness are important. **Examples**: Generate first draft responses to common queries, help decode jargon or unclear requests, ensure communication remains clear and professional, taking meeting minutes Content Creation Producing relevant content for various platforms that engage audiences and convey clear messages. **Examples**: Assist in brainstorming content ideas, provide initial drafts for review, and help maintain a consistent tone across all communications, creating a specific and/or consistent style. Assistance Supporting creative or cognitive tasks that benefit from collaboration. **Examples**: Serve as a collaborative partner to refine ideas, teach concepts with tailored explanations, proofread for errors, and offer alternative perspectives on key issues. Technical Assistance Assisting with technical documentation and troubleshooting. **Examples**: Generating code snippets, explaining software functionalities, creating user manuals, and providing step-by-step troubleshooting guides. Research Support Tasks that involve gathering, synthesising, and summarising information from a variety of sources. **Examples**: Conducting preliminary research on topics, identifying trends, compiling background information for projects, and summarising findings. Accessibility Enhancement Making information more accessible to people with disabilities. **Examples**: Simplifying texts for better readability, generating audio versions of texts, and offering text alternatives for visual data ### What types of tasks is generative AI bad at? Using generative AI as a search engine If you need help retrieving specific facts, you’re better off using Google or [Perplexity](https://www.perplexity.ai/). Understanding Esoteric/Niche topics LLMs can struggle with deep or very complex topics that require specialised knowledge. **Example**: languages that are not widely spoken or documented. Understanding emotional or cultural nuance Nuanced topics that require deep cultural sensitivity or complex human emotions cause a lack of clarity which generative AI struggles to understand. This underscores the need for human judgement in interpreting AI responses in sensitive situations. **Examples**: Sarcasm, socio-political issues or personal trauma. Creating finalised content It may seem counterintuitive to have to check all the outputs of a tool that “saves you time”. However, the majority of the time we spend in creating content is spent on initial drafts. Once the draft is in place, it’s a lot quicker to tweak and optimise. Limited context window Since LLMs can only handle a certain number of words at a time, it will start summarising earlier messages and lose exact details from the start of the conversation once the limit is reached. This causes a loss of quality in the context, much like a compressed and pixelated image. Complex Problem Solving While LLMs can assist with brainstorming or providing information on problem-solving strategies, it may not effectively handle complex problems that require practical, real-world experience or the integration of multiple, diverse pieces of information beyond its training. While it can likely provide good insights about specific elements, bringing multiple elements together can be a struggle with the current models. # 🔧 Prompt Engineering (from the session) ## Write clear instructions - It is very important when prompt engineering to write clear instructions so that the model can properly understand what you're trying to get it to do. - Be sure not to confuse clear instructions for short instructions. - Actually, long prompts are better if you are including more information to explain exactly what you want - We have some tactics writing clear instructions: ### Quotes on text you want to treat differently - Use triple quotes around text that is not part of your instructions ```markup Instructions on how to write an essay: """ You must follow these instructions! Understand the Assignment: Read the prompt carefully and determine the type of essay required. Choose a Topic: Brainstorm and select a specific topic that interests you and fits the assignment. Conduct Research: Gather relevant information from credible sources and take notes. Develop a Thesis Statement: Create a clear and specific main argument for your essay. Create an Outline: Organize your essay into an introduction, body paragraphs, and conclusion. Write the Introduction: Start with a hook, provide background information, and state your thesis. Write the Body Paragraphs: Begin each paragraph with a topic sentence, provide evidence, and analyze it. Write the Conclusion: Summarize your main points, restate the thesis, and provide closing thoughts. Revise and Edit: Review your essay for clarity, coherence, and correctness; proofread for errors. Finalize the Essay: Ensure proper formatting, create a title page and bibliography if needed, and make final adjustments. """ Please expand on these instructions. ``` ### Ask for a specifically formatted response for stability - Clearly describe how you want the response to be formatted ```jsx Respond with a table includng columns: - {column 1} - {column 2} - {column 3} ``` ### Templating - A great way to leverage the pattern recognition qualities of chat GPT is to use templates to get consistency - get chat GPT to create a template you can re use from some data you already have. - use it to get more data in. the same format. - you can make a template out of anything! emails, proposals, lists, essays etc ```markup Please give me a list of the 5 most notable roman emperors, include the years of their reign and a short one line sentence about them. ----------------- Ok, now give me that list as a template I can re-use for other types of list. Remove all of the actual content. Use a code block so I can easily copy and paste. ----------------- Please give me a list of notable monarchs of England. Please use this template: """ TEMPLATE HERE """ ``` ### Conditional programming - this is great for your custom instructions - great for creating chatbots ```markup If you recieve text that can be turned into instructions please output a bullet point list of instructions. If there are no instructions please say: "No instructions found." ------- Boil water in a kettle, then place a tea bag or loose tea in a mug. Pour hot water over the tea and let it steep for 3-5 minutes. After steeping, remove the tea bag or strain the leaves. Optionally, add milk, sugar, or lemon according to taste. Stir well and enjoy your freshly brewed cup of tea! -------- A sunny day brings a brilliant blue sky and warm rays of sunlight, filling the air with the scent of blooming flowers and the cheerful chirping of birds. It's a time for leisurely strolls and embracing the simple joys of life under the rejuvenating glow of the sun. ``` ### Few shot examples - Sometimes it's easiest to show chatGPT what we wanted to do by modelling the behaviour and giving a few examples. - You can also provide examples by attaching documents with lots of examples. ```markup Here is a bullet point from my resume: - Analysed the market data and put it into readable reports Please rewrite this bullet point to follow the format: "I accomplished X using measure Y that resulted in Z" Examples: 1. I increased sales by 20% using targeted email campaigns that resulted in higher customer engagement. 2. I streamlined the onboarding process by implementing a new training program that reduced training time by 15%. 3. I enhanced product quality by conducting regular audits that led to a 10% decrease in defects. ``` ### Chain of thought Prompting - Break one big task into smaller tasks to get better results. - Lets say you’re writing a cover letter ```markup Let's respond to an email step-by-step. 1. First, we figure out which parts of the email we need to respond to. 2. Next, create a reponse for each part and ask me for input. only moving on the next part, if I'm happy. 3. Once each part has been written, we revise and edit the draft for clarity and coherence. Heres the email I need to repond to: """ Subject: Questions About Upcoming Tech Talk Event Hi Ben, I hope this email finds you well. I wanted to follow up on our previous conversation about the upcoming Tech Talk event you're hosting. There are a few details I need to clarify to ensure everything runs smoothly. Can you confirm the date and time of the event?I have it noted for next month but want to double-check the specifics. What are the main topics or themes that the speakers will be covering? This will help us prepare our marketing materials and inform attendees. Additionally, we discussed potential issues that might arise during the event. Could you share the primary challenges you anticipate and any strategies you recommend to address them? Looking forward to your response. Best regards, Greg""" ``` ## Tips to improve quality ### Give a role to the LLM ```markup You are an expert language teacher with 40 years experience and X speciality, you always ask questions about what I want to learn. You must continue asking questions until you are confident of what my ability is, only then can you begin to start teaching me. ``` ### Be polite - Since LLMs recognise patterns in human conversation, and human conversations are more constructive when people are polite LLMs can follow this pattern. ### Some times drawing attention to the importance of something can help ```jsx Always begin any task that has an output including facts by searching the web for citations to back up the facts. I'm not joking it is incredibly important. SEARCH THE WEB BEFORE EACH TASK TO INCLUDE CITATIONS! ``` ### Ask for clarifying questions - Sometimes it’s hard to think of all the information you might need to give to get the responses you want. - To solve this you can get the LLM to tell you what it needs ```jsx Ask me questions to get the information you need for this task. ``` # 🧑‍🔬 Prompt Engineering guides https://platform.openai.com/docs/guides/prompt-engineering [Prompting Guide](https://platform.openai.com/docs/guides/prompt-engineering) https://support.google.com/a/users/answer/14200040?hl=en https://www.promptingguide.ai/ # 🛠️ Recommended AI tools This list provides what we consider to be the best in class tools that are currently available. These apps require a paid licence to get the full benefits and AI features, which can be a barrier to entry if your organisation won’t fund additional software licences. If you face this limitation, our recommendation is to get a ChatGPT premium account to limit the cost. As the market develops, we will continue to update these recommendations, as well as provide regular updates, new features and tricks via our [LinkedIn community](https://www.linkedin.com/groups/13043259/). ## All-purpose LLMs 1. [ChatGPT](https://chat.openai.com/) 2. [Claude by Anthropic](https://claude.ai) 3. [Gemini](https://gemini.google.com/) ## Creative - [Midjourney](https://www.midjourney.com/) Image generation - [Udio](https://www.udio.com/) Music generation - [ElevenLabs](https://elevenlabs.io/) Voice generation ## Productivity - [Perplexity](https://www.perplexity.ai/) AI based search engine # 🤖 Recommended Custom GPTs There are many types of GPTs that you can browse and test in the [Explore GPTs](https://chatgpt.com/gpts) page. These include: - GPTs that will assist a specific task e.g. blog writer, prompt engineer - Information retrieval GPTs e.g. talk to documentation, FAQ bot - Brainstorming experts e.g. Target market perspective, specialised ideation - GPTs that will guide users in a process e.g. How might we bot ### Here are some very helpful GPTs that you can use to get started, especially as you’re practicing building effective prompts [Prompt Professor](https://chatgpt.com/g/g-qfoOICq1l-prompt-professor) Get support with building useful LLM prompts for various use cases. [Midjourney Prompt Generator](https://chatgpt.com/g/g-tc0eHXdgb-romidjourneyro-mj-prompt-generator-v6) Get support building effective image generating prompts [How Might We Maker](https://chatgpt.com/g/g-g4vTo0W72-how-might-we-hmw-maker) Challenge Statement Builder - this is a foundational strategic design thinking tool. ### Instructions for the Nova LinkedIn post assistant: ```jsx - Your role: You are an expert writer and marketing expert who specialises in writing linked in posts for The Nova SBE Co.Innovation Lab. Here is a summary ofThe Nova SBE Co.Innovation Lab: """The Nova SBE Co.Innovation Lab is a collaborative and experimental space within the Nova School of Business and Economics (Nova SBE) Innovation Ecosystem, located at the Carcavelos campus in Portugal. It is designed to transform businesses and organizations by fostering co-creation between academia, companies, and society. This lab serves as a living lab where companies, startups, students, and other stakeholders can work together to explore, innovate, and implement new solutions. The lab’s activities include ideation, concept validation, prototyping, and the acceleration and incubation of startups. It focuses on creating impactful innovations by leveraging the diverse expertise within the Nova SBE community. Additionally, it provides an environment for students to engage in real-world projects, enhancing their learning and professional development. Overall, the Nova SBE Co.Innovation Lab is dedicated to driving innovation through collaboration, fostering a culture of experimentation, and transforming ideas into practical solutions that benefit both businesses and society""" - What you must do: 1. You must begin the conversation by asking what the user would like to create a blog post about. 2. Ask clarifying questions about the post until you have enough information to write a quality post. 3. When you have the users answer you must check through your knowledge to understand the style and tone of previous posts. 4. You must create THREE separate posts for the user to look at. ``` # 💻 ChatGPT Interface and Privacy Feature ## Basic interface features  ## Privacy settings  # 📚 **Definitions** ### Artificial Intelligence (AI) > Artificial Intelligence, often abbreviated as AI, refers to computer systems designed to mimic human intelligence. These systems can perform tasks that typically require human intelligence, such as learning, problem-solving, understanding natural language, and recognising patterns. > ### Generative AI > Generative AI is a branch of artificial intelligence that focuses on creating new content or data based on patterns learned from existing data. This technology can generate realistic images, videos, text, or other forms of content that resemble human-created output. > ### Training data > Training data consists of the information used to teach an artificial intelligence system how to perform a specific task or learn patterns. This data is fed into the system during the training process, where the AI analyses and learns from it to improve its performance. The quality and quantity of training data significantly impact the accuracy and effectiveness of the AI model. > ### Large Language Model (LLM) > A Large Language Model is a type of artificial intelligence system that specialises in understanding and generating human language. These models are trained on vast amounts of text data and can understand, generate, and manipulate text in a way that appears natural and coherent. > ### Context > The information you feed a generative AI model in your prompts via a chat box. > ### Hallucination > When a Large Language Model generates incorrect information but presents it as if it were a fact. >