Baby Steps

Jeremi recaps a project breakthrough; Luca writes about learning in public

Welcome back to Jeremi and Luca’s Newsletter, a weekly update from two friends connected by a relentless desire to learn.

Enjoy a closer look at what we’re reading, working on, and thinking about—from the small learnings to the large perspective shifts.

---

Jeremi: Stepping in Simulation

By the time it hit 8pm, I had already admitted defeat.

My morale had been the lowest that morning, as I felt the full disappointment of not being able to complete the one part of this project that I wanted.

Halfway through this quarter, I joined a graduate-level class on Robot Learning. I initially just expected to crash the course, but was able to become enrolled after some convincing of the school admin and the professor.

If I were just interested in the content, crashing the lectures and taking notes could have been enough to satisfy my curiosity. But the main part of the course, the project component, was something that I wanted to be enrolled for. Sure, I could have followed along with my own personal project (my plan if I didn’t get in). But being enrolled forced me to prioritize the project more than I might otherwise, with deadlines and teammates to collaborate with.

From the moment I joined, it was a race against time. I joined a group with only four weeks left to go in the quarter, and they pivoted during the same week. We were starting fresh.

I don’t think of myself as a big procrastinator, and I don’t think I procrastinated, per se. I worked on this project, made progress when I could. And yet, I knew it wasn’t enough. In the last week, I was looking at a project my team was seriously behind on, in a class the professor had made an exception to allow me to join. This was bad.

Thus began the most intense week of the quarter. I put everything I had into the project in this final week. Our project was based around humanoid robots walking in simulation. My role was to build an algorithm from scratch, which had one goal: make the robot walk. This was a requirement the professor had been clear on—we needed to implement at least one of our machine learning models ourselves.

***

Monday night at 8pm, 50 hours of work and a triathlon later, I had come to terms with the fact that we wouldn’t accomplish our goal. The robot didn’t walk, and our presentation was the very next day at 11am.

I was disappointed after having put all my effort over the last week on this, only to come up short. I was worried about how the professor would judge me and our project.

By 8pm, these feelings had faded, and I began documenting my work and making plots of the performance of the model under different conditions. I had lost hope of any final miracle.

Until the miracle happened.

By documenting the performance, stripping each component of the model down, and evaluating the change in performance when I added that component back, I noticed a subtle issue in the code that had blocked the model from learning effectively.

I didn’t expect it to make that big of a difference. I retrained the model, and the visualizer showed a robot making awkward, stiff, but true steps. I leaped out of my seat. “It’s walking!” My teammates could hear the joy and relief in my voice. I was in disbelief. I kept repeating those words, over and over: “It’s walking. It’s walking.”

It has been a pretty stressful week—multiple 16 hour days of working on this project while trying not to miss my last lectures and prepare for finals.

But even in the final hour, I didn’t regret it. The looming consequence of failing on our project pushed me to work harder than I ever would have otherwise. When it looked the bleakest, I had come to terms with it—I had learned a lot and had let go of what the outcome of the project would be.

Funny how that’s exactly when a miracle struck.

Update on the triathlon: beautiful race with amazing people!

---

Luca: Stepping in Public

I was texting my friend Jed this week and we were discussing SF events. He’s a recent college graduate who works in the startup banking arm of J.P. Morgan.

Jed had this to say:

Jed was referring to the rise in popularity of people talking about what they’re building (usually tech) rather than staying in “stealth mode.”

In effect, people seem to be trading strategic secrecy for strategic publicity.

Why? Presumably, many reasons: accountability, feedback, idea refinement, exposure to investors and other supporters, community, etc.

Some of my favorite people to follow online are those who “build” in public. Paul Graham writes regular essays about what he’s learning. David Senra’s Founders Podcast is just him talking about the insights he gains while reading biographies.

The list goes on and on, and the practice is only growing in popularity.

It’s why Jeremi and I write this newsletter: we’re effectively “learning in public,” keeping ourselves accountable to learn something new every single week.

Everything we write here has positive value; either we deepen our understanding of topics through writing or we receive helpful feedback and ideas from our readers.

And ideally, we hope that what we write can add some value to you all as well.

At a higher level, I’ve enjoyed the “building in public” benefits of what I’ll call a “newsletter planning” mindset—that is, our weekly update serving as a requirement that I do something interesting each week. It keeps me at a fast pace, zooming in on groups of seven days as recurring opportunities to be stringent with my time.

Conversely, a heuristic: if I ever get to Sunday and have nothing to write about, my week was probably not very worthwhile.

And thus, why we’re learning in public.

Mid-semester wind ensemble concert last night