Talks I Love: STEM For General Audiences

This is part of the Talks I Love series.

For this month I wanted to branch out and focus on technical or scientific talks from outside of the Ruby community. I found both of these on a list of STEM focused TED Talks. I have watched most of the talks, and they are all great. I picked these two for this post because they do a fantastic job of making hard mathematical concepts concrete and understandable by a lay audience.

When I interview folks for Developer Relations one of the skills I try to assess is whether the candidate can tailor their message to the audience. More skilled people understand deeper explanations, technical jargon, and analogies to other technologies. Less knowledgeable audiences need to understand the big picture. They benefit from comparisons to everyday objects and experiences.

Both of the speakers I highlight this week do a great job of explaining the challenge and promise of their respective fields in a way that is approachable without speaking down to the audience. Adapting your message to your audience is a skill that I believe all technical folks should nurture and value.

Fei-Fei Li: How we teach computers to understand pictures

This talk starts with a concrete and relatable example, a child describing photographs. This is something many of us can relate to instantly. We probably don’t even think of this as something that we teach to children. In using this example, Fei-Fei helps us understand that computer minds and human minds have different strengths and weaknesses. She also tells us how improving computer vision could make the world a better place. Also, she does all of this in less than three minutes. It is an incredibly strong introduction, and I want to try something similar in some of my talks.

Throughout the talk, Fei-Fei uses visuals to help the audience follow. When discussing how computers represent images she overlays a grid of numbers on top of a picture. Showing a sitting cat and superimposing a stick figure on top of it helps the audience understand the old approach. Then by showing some additional examples, a cat curled into a ball, a cat hidden behind an object, and a cat on its hind legs she shows how that model fails. When she is discussing the new approach that teaches the computer with millions of images she shows a slide with dozens of cats of all breeds and species linking the examples together but also help make the differences concrete.

At the end, Fei-Fei comes back to the original example by having the computer describe the same images that the child described in the beginning. She again shows the limitations of the computer by giving some examples it does not accurately describe which helps the audience both understand that the system has limitations but also makes the demo more believable. After that, she again goes through some ways that improving computer vision can help humanity.

As I was watching this talk, I was thinking about the advances in computer vision that Google has integrated into products over the last few months. This talk was given about two years ago, and last month at Google I/O it was announced that Google Lens is better than humans at recognizing objects in photos. In March, at Google Cloud Next, Fei-Fei Li, now a Chief Scientist at Google, announced Google Cloud Video Intelligence which does computer vision analysis on videos instead of still photos. Computer vision and machine learning have come so far in such a short time.

Hannah Fry: The Mathematics of Love

First off, Hannah’s delivery is upbeat and amusing. Both her delivery and the topic challenge the misconception that mathematics is stodgy and boring. She also has an introduction that quickly gets to the point and makes the content of the talk relatable. She starts with a paper called “Why I Don’t Have a Girlfriend” and accompanies her speaking with a visual that eliminates stick figures for different reasons at each step of the mathematical analysis. This visual lets the audience follow along and also is the setup for her joke about there being more extraterrestrial life forms that eligible women for the paper’s author.

When she discusses the OkCupid data, she again uses visuals to help the viewers understand what the mathematics is saying. She shows the graph from OkCupid which has a roughly up and to the left correlation. I saw the chart and concluded, “the more attractive you are, the more messages you get on dating websites.” Hannah points out that is the wrong conclusion to draw. To help folks understand the correct conclusion Hannah uses pictures of two celebrities to show that the data is telling us that people who get a wider range of “attractiveness scores” will get more contacts than people with a smaller range. The example links the abstract nature of the graph to something that the audience can see and understand so that the chart and the conclusions drawn are easier to understand.


The “Talks I Love” series about speaking will be going on hiatus for a while. I enjoy writing these posts. I believe that speaking is a craft that is just as worthy of study and refinement as programming and other technical topics. Even within developer relations we often treat what we do as somehow less than software engineering. However, being a good communicator is complicated, multifaceted, and important. In the future, I hope to do another series focused not only on speaking but on Developer Relations skills in general. I want to discuss writing abstracts for CFPs, building community, scheduling content, advocating within your company for your users, and all the other skills that are under the giant umbrella of DevRel.

In place of “Talks I Love”, the first Thursday of every month will be a post focused on the very basics of machine learning. Machine learning has always been intriguing and somewhat mysterious to me. To me, it feels like the math and programming involved should be beyond mere mortals. As I found out in an Artificial Intelligence class in college, AI and ML are mostly just more programming. There are some rules of engagement for analysis that are borrowed from statistics, and there is some linear algebra. The basics, though, are within the ability of anyone who managed to get through a semester of college math or the equivalent. I will be coming to the ML posts as a hobbyist speaking to other hobbyists, so I hope the material will be approachable. We will see how I do the first week of July.