- EASie AI
- Posts
- Issue 1 AI Literacy for Parents and Educators
Issue 1 AI Literacy for Parents and Educators
Educational App Store
June 11, 2024
Welcome, cutting-edge educators and aspirational parents
Introducing your no-nonsense EASie guide to using AI to support your kids’ learning and growing your own AI literacy
Today, we’ll cover:
Automated teaching resource creation tool
Free math solver inbound to your iPad
AI jargon buster [beginner] - LLM
AI jargon buster [advanced] - Training
AI Fact of the Week - Why AI is an E- in math!
AI Safety - Super Teacher’s approach for kids
AI News - quick updates
Total reading time: five minutes
Tool For Home Educators and Teachers
Source Dall-E
When an instructional designer makes a tool, you know it will be intuitive to use. Learnt.AI has loads of standalone AI tools, which makes it easy for you to choose your task and get your generated educational material.
Step-by-step
Go to Learnt.AI and click Try Free at the top of the page.
Create a free account
Choose ‘All Tools’ once you are on the dashboard, then look for the orange open padlocks indicating free tools.
Choose the tool and enter your requirements
Sit back and let the AI generate courses, analogies, homework tasks and more for free!
For iPad Users
A calculating mind!
Have you ever noticed the lack of a calculator on your iPad? Soon, the void will be filled when Apple releases its Calculator app for iPads. Apple announced at its conference on Monday that the Calculator will have AI math-problem-solving capabilities.
Parents should keep an eye on this, as Solver apps can undermine education. It seems like the Apple one is all about getting solutions. It sounds great for day-to-day productivity, but not so much for education.
AI Jargon
Source Dall-E
We’ll help you increase your knowledge of AI by explaining a beginner and advanced concept each week
For AI beginners
LLM
Let’s start with the one appearing on all the news. An LLM, or Large Language Model, is a computer program that can understand and generate human-like text.
Trained on huge amounts of data (books, websites, text, documents) over a long time
LLMs generate their output by predicting what is most likely to follow what came before
They add a little randomness to the prediction to sound more human at the expense of making more mistakes
The question or task you enter into an LLM is a prompt.
LLMs use multiple prompt types – yours is the user prompt, but there are also system prompts which set the LLM’s constraints and safeguards
Red teaming is a testing process where researchers attempt to get LLMs to break the rules set by the LLM designer or app in the system prompts.
For AI Adepts
Training
Training an LLM takes a long time. The algorithm goes through the entire dataset and tries to predict the next word (or token) in a sequence. If it gets it wrong, it changes its parameters and tries again until it gets it right. It gradually becomes better at predicting the next token.
The process of going through a whole dataset is called an epoch. Training an LLM requires multiple epochs.
Too few epochs result in an inaccurate LLM.
Too many epochs result in ‘overfitting’ where an LLM can only regurgitate its training and not respond to new information
Fine-tuning helps LLMs become better at specific tasks and roles.
People often talk about training an LLM by providing more information, such as text in PDF format. This is called Retrieval-Augmented Generation. While useful, it is not true training and is not as effective but it is much more flexible and quicker.
AI Fact: AI Can’t Count!
E- At Math!
LLM’s are terrible at math. In all their training data, they’ve probably encountered some simple arithmetic like 5+5=10, so they will (probably) get it right.
However, seeing 34234342+4356345 is much less likely, and as they have no concept of counting, they can’t do it.
Except…they can. Modern LLMs detect when you ask a math question and pass it on to a subroutine that calculates, not predicts the answer. It’s like you reaching for a calculator rather than doing some mental math!
AI Safety
Best Practices in AI for Education
Source Dall-E
Generative AI is considered risky for young kids because it produces material that has never been approved by humans. With infinite possible prompts, we can never be sure that generative AI will not present something undesirable to one of them.
Super Teacher avoids generative AI but uses other AI types, such as Natural Language Processing, to listen to kids, understand what they say, and present pre-approved material to match.
Read their design process to see how it uses AI to support young kids while staying on the side of safety. Super Teacher Design Process.
AI News
Your one-stop, in-brief AI news stand
Source Dall-E
Apple announced many new AI-based features for its products, but you might need to upgrade to use them. For example, many AI features for the iPhone won’t work on anything less than the iPhone 15 Pro.
Privacy and AI continue to conflict as Microsoft reels back on its plans for Recall. This technology essentially records everything you do on your PC so you can use AI to find previous files, images, and websites with natural language.
Elon Musk cited privacy concerns when he threatened to ban Apple technology in his companies if Apple builds Open.AI technology into its devices at an operating system level.
Irresponsible apps threaten to undermine the whole technology, and Google introduced new rules for publishing its developers on Google Play. Every generative AI app must allow users to flag inappropriate responses.
We hope you’ve learned something new about AI and that we’ll see you again. When used safely, responsibly, and effectively, AI holds great promise for education. This newsletter is one part we want to play to help raise AI Literacy skills for parents, teachers and kids.
We have other plans to help ensure AI is a force for good in education—you’ll find out more about those in future newsletters. See you then!