Writing explainers with the 1000 most common English words
This semester I decided that I wanted to improve my writing a LOT. I want to communicate better about science and, honestly, everything. Right now I’m doing this NYU workshop about writing for a broader audience and we had a very fun exercise to do: we had to write an explainer using only the most commont 1000 English words. We used this website that would turn all the “forbidden” words red.
So, you might be wondering what an explainer is. I got you: it is a piece of text that explains in simple words something. This “something” can be a concept, a tool, a method and so on. In my case, I decided to explain three different machine learning training paradigms: supervised, unsupervised and self-supervised learning.
I’m sharing here both the original and editted versions because I recently understood that no written text will be perfect from the very first draft, even if you do a good job. We all go through rounds and rounds of editting to have a final nice piece of work. So don’t give up. Keep writing!
Submitted Version
Let’s say that you want to teach your computer how to recognize a dog. I’ll tell you three ways of doing so.
The most usual way of doing it is to give your computer a picture of the dog, and then tell it, “This is a dog”. You repeat this a couple of times with other dog pictures, so your computer understands what makes a dog… a dog. In this case, the computer just learn what you decide to teach. If you choose to teach about dogs and then ask it about a bird, you cannot expect the right answer. This is a new animal! You can even choose to confuse your computer if you want: just give it a lot of dog pictures and tell it they are all cats. The computer will repeat this forever and say that every dog is, in fact, a cat. Your computer will not fight you and tell you that you are wrong; it will just learn whatever you teach it (right or wrong).
The second way of teaching your computer things is to give it a lot of different animal pictures (you name it) and let it figure out how to group them. As you can imagine, there are different ways of grouping the animals: by color, by size, by animal, and by all of these together. As the computer chooses how to group the images, it learns what the most important things are about that group and what makes that group different than the others. These things can be later used to learn more about newer animals if needed. If you have never shown your computer a bird picture, it will still be able to understand that it looks different from a dog (because a dog does not have wings). This is not to say that it will be perfect: the computer may group the bird with that other animal with wings. In this way of learning, your computer will be able to understand what makes a group a group and what is the closest group to the animal you’re asking about, but it won’t know what the animal’s name is.
The third way is the two together. You start by asking your computer to group animals, and it will learn the important things about each group. Once your computer understands that, you give pictures and names of the animals you want the computer to recognize: dogs, birds, cats, and so on. Because your computer knows important and different things about animal groups, it will take less effort to learn what is so important and special about the animals you are interested in.
Edited Version
Let’s say that you want to teach your computer how to recognize a dog. I’ll tell you three ways of doing so.
The most usual way of doing it is to give your computer a picture of the dog, and then tell it, “This is a dog”. You repeat this a couple of times with other dog pictures, so your computer understands what makes a dog… a dog.
In this case, the computer just learns what you decide to teach. If you choose to teach about dogs and then ask it about a bird, you cannot expect the right answer. You can even choose to confuse your computer if you want: just give it a lot of dog pictures and tell it they are all cats. The computer will repeat this forever and say that every dog is, in fact, a cat. Your computer will not fight you and tell you that you are wrong; it will just learn whatever you teach it.
The second way of teaching your computer things is to teach it to make groups of similar things. Here, you give it a lot of different animal pictures and let it figure out how to organize them. There are different ways to do it: by color, size, animal, or all of these together. As the computer chooses how to group the images, it learns what the most important things are about that group and what makes it different than the others. These things can be later used to learn more about new animals. If you have never shown your computer a bird, it will still be able to understand that it looks different from a dog: one has wings and the other doesn’t. This is not to say that it will be perfect: the computer may group the bird with other winged animals. With this approach, your computer may understand what makes a group a group, but it won’t know any animals names.
The third way is to combine the other two approaches. You start by asking your computer to group animals, and it will learn the important things about each group. Once your computer understands that, you give pictures and names of the animals you want the computer to recognize: dogs, birds, cats, and so on. Because your computer knows important and different things about animal groups, it will take less effort to learn what is so important and special about the animals you are interested in.