Work with generative AI

1. Prompting, chatting with AI
 

Prompting is a way of guiding a conversation with chat-based AI tools. It's like a cue or a nudge that guides the AI to respond to your questions or perform tasks. 

Example of prompting with AI tools, in this case Aalto AI Assistant:

"Hello, please improve my english, <your text>"

These prompts could be in the form of sentences, questions or phrases that instruct the AI. Think of it as giving instructions to a friend - you tell them what you need, and they do it for you. Similarly, your interaction with AI through prompts helps it understand and deliver the intended outcome.

Your ability to write prompts will improve over time; however, here are some tips and examples to help you along the way:

Iterate and refine. Start from the big picture to get generic ideas and points of view. Then dig deeper with precise angles and examples.​

"Please list phenomenon related to sustainability, from the fashion industry's point of view."

Provide examples. Provide examples of the tone, vocabulary and other elements you’d like to see in the response.​

Always review and edit results before use​

• Language model is not perfect and often provides biased or wrong outputs and facts​
• You are responsible for validating AI results before using them​
  
   

2. General examples of prompting

Grammar Check: Polishing Your Text

“Please improve my English, <add your text here>”.  Tip: After the answer you may simply say if you want easier English. 

Changing the style of language

“This text, below, if written by a <topic> professional, the style is too complex for a non-professional reader.  Please rewrite the text, by using everyday wordings.”

Brainstorming

• “I’d like to introduce a <topic>, from <the point> of view. Please suggest me 10 examples, how to say it.”

• “I need ideas for <your topic>. There should be at least the following details and point of view: <your  details”.  

  Tip: You can always let AI Assistant rewrite any ideas but this time with another angle or format.

Text comparison: with suggestions what to add

“I have two texts, below. 
The first one is a bit longer, with more examples and details. 
The second one shorter but also lists topics in a different order. 
Please compare the texts and feature the possible differences.  
Is the second one missing any major topics?

<Your text pasted here>”

Tip:  If your text is long (longer than 2 pages of A4 papers, full of text, and the answer will be another 2 pages of A4 full of text), please change the language model of AI Assistant to "GPT ... long document". This way your input and AI Assistant's output may be longer, appr. 20 page of A4 papers

Multilingual content production

• “I will have 10 text paragraphs below.  Translate them to Swedish, Finnish, Norwegian, Russian and French”
  or      

• “Please create a table of vocabulary: I will list 50 words, make it the first column, and then add Finnish, Swedish and Russian translations as the next columns.”  

  Tip:  Especially the vocabularies need context, because one word may have many meanings. You may add the context next to a single word , “net (of strings or ropes)”, or you may give the context for all of the following word, “All the words relate to fishing and hunting”.

Creating meeting memos

“We recorded a Teams meeting and downloaded a meeting transcription of it, see below. Please make a meeting memo with the following topics.  

1 - List of participants, based of who talked during the meeting
2 - List of topics, only the major topics
3 - List of action points or tasks
4 – A bonus: Someone in the meeting promised to do the task <described here>. Please read the discussion and who said he/she will do it? "

Creating questions and quizzes

• “My text is below. Please create question + answer pairs of it.”    
  Or
• “My text is below. Please create a quiz with 20 questions and 3 answer alternatives for it. Mark the correct answers as a list in the end.”
   

3. Who are you chatting with: What is Large Language Model (LLM)?

When you are chatting with AI, like OpenAI's ChatGPT service or Aalto's AI Assistant, you are chatting with Large Language Model (LLM).  These models are designed to understand, generate, and work with human language.

Language models predict the next word, based on patterns and context, see below.  Language models are trained  on a large amount of text and data to understand how people communicate. 

4. Limitations of Large Language Models

Inaccurate, unreliable or incomplete information: LLMs are word generators. Do not assume that the answer given by any language models is accurate or reliable. Therefore, please fact check the answers.

Effectiveness can vary based on how the prompts are phrased and the complexity of the topic -> Practice with prompts and use prompts based on CLEAR principle (Concise, Logical, Explicit, Adaptive, Reflective).
Biases are inherited: LLMs inherit biases present in their training data. They are designed to output what has most frequently been seen, not what should be the correct answer to your question.

Contextual limitations: LLMs do not fully understand the nuance or context of complex issues. They can generate plausible-sounding but incorrect or irrelevant answers.

Limited creativity: LLMs struggle to produce truly original or imaginative content, relying instead on patterns learned from training data.

Out-of-domain responses: LLMs provide inaccurate or irrelevant responses when faced with queries or prompts outside their training domain, as by default they still try to provide some answer even if they don’t have enough information to provide an answer.

Lacking consistency and coherence: LLMs may exhibit inconsistencies or lack coherence in their responses, especially in longer or more complex interactions. This can result in disjointed results.