The AI Playbook

Cheery Friday Greetings to our Learning How to Learners!

Book of the Month (to be released February 6th!)

The AI Playbook: Mastering the Rare Art of Machine Learning Deployment, by Eric Siegel. Whether you are an AI expert or know nothing about AI, this book will teach you, through compelling examples of success and failure, what machine learning projects are and how to implement them.  Siegel is a master story-teller, who starts us out by describing how the delivery company UPS implemented machine learning to streamline its business practices.  This may seem like a no-brainer, but at the time, economizing with the help of machine learning was pretty much the last thing on the mind of most UPS business executives.  Indeed, it’s easy to go off track with machine learning projects, which aren’t like simple plug-and-play computer products, but instead can involve entire swathes of a company’s different divisions. Through the corresponding sections of his book, Seigel guides us through what he terms “BizML Practice”:

1. Value: Establish the deployment goal
2. Target: Establish the prediction goal.
3. Performance: Establish the evaluation metrics.
4. Fuel: Prepare the data.
5. Algorithm: Train the model.
6. Launch: Deploy the model.

Seigel’s broad experience in every aspect of the field helps bring the practices he describes to life.  We LOVED this book and cannot recommend it more highly!

Make class prep easy with 8 flexible prompts for retrieval practice

We’re frequently asked for tips on how to encourage and use retrieval practice in classes.  This marvelous article by Pooja Agarwal provides simple, practical ideas that you can put right to work in your teaching.  And of course, don’t miss the book Powerful Teaching, which gives even more resources.  Enjoy!  

How Transformers Helped AI Vastly Expand Its Grasp of Language and Meaning

Many different approaches have been tried over the years to allow a computer to rapidly parse the relationships of all the words of a sentence, paragraph, book, or in fact, any consecutive string of information. Computers’ ability to hold and analyze large chunks of information at once is akin to the abilities of a human’s “working memory”—that is, how much information we can grasp and hold in mind at once. (Please forgive us for attributing humanlike qualities here, but giving human traits to AI makes it easier to grasp how ChatGPT and comparable advanced natural language models work.)

In previous decades, we humans have experienced the upshot of computers’ limited computer working memory when it came to language translation. At first, computers could only hold only a single word in mind when translating into a foreign language.  This resulted in strange translations like “The test is a piece of cake,” (meaning “the test is easy”), being literally translated into Spanish as “El examen es un pedazo de pastel.” (Really? Should we eat the test, then?)  

In contrast to computers, human can hold up to an average of four “chunks” of information in mind at once. Those chunks can be quite large if they’re connected to information stored in long-term memory.  (Retrieval pratice helps here!) This means it’s straightforward to hold the sentence “The test is easy” in mind if you are an English speaker—you’ve got a lot of knowledge of English semantics, grammatical structures, and writing stored so you can retrieve the information without even thinking about it.  But just try holding the same sentence in mind if you haven’t learned Russian: “Экзамен лёгкий.” Or you don’t know Chinese: “考试很简单.”

To expand what a computer can process in one go, scientists tested different methods over the years, like recurrent neural networks and gated recurrent units (types of neural networks). Computer translations improved markedly when computers began to be able to hold entire sentences, and then paragraphs “in mind.”  But matters become more and more difficult the larger the strings of words become—it takes a lot of computational horsepower to remember everything that was written, for example, in the first half of a book.  

Large Language Models exploded into the public eye when Google engineers found a way to make for near-infinitely large computer working memories that can allow computers to hold entire books, and more, in their working memory equivalents.  These new Large Language Models use what are called transformers.

“Attention is all you need,” the paper that introduced the concept of transformers, has been cited over 100,000 times since it came out in 2017.  But despite its importance, this paper can be hard to parse.  This 15-minute video by the AI Hacker explains the operation of the transformer in as simple and clear a set of terms as we’ve seen anywhere. If you’re trying to understand ChatGPT, and you have a little bit of a tech background, you’ll enjoy!

A terrific direct eye-contact webcam

And speaking of attention, David Joyner, our co-instructor for our marvelous (even if we do say so ourselves!) course on “Teaching Online,” recommends the “IContactCamera.”  This allows your viewers to focus and pay better attention to you.  David writes: “The stick that holds the camera out is way more stable, the camera quality itself is a good bit better than the CenterCam (not quite to the level of the Brio, but only really notable in difference when comparing them side-by-side), and I like how easy it is to flip it out of my way when I’m not in a meeting or using a teleprompter.”  

That’s all for now. Have a happy week in Learning How to Learn!

Barb, Terry, and the entire Learning How to Learn team

• Uncommon Sense Teaching—the book and Coursera Specialization!
• Mindshift—the book and MOOC
• Learn Like a Pro—the book and MOOC
• The LHTL recommended text, A Mind for Numbers
• For kids and parents: Learning How to Learn—the book and MOOC. Pro tip—watch the videos and read the book together with your child. Learning how to learn at an early age will change their life!