RC Retrospect

We have the hub in Brooklyn. This is the physical place of RC. But RC is so much more than that, and it's also not static, but a changing living entity. Certain things change more slowly, like the hub and the tools we use together with the spectacular and kind admins. Other things do change rapidly (almost, like, every 6 weeks…¹). Being out of batch and still going to the hub it feels.. different. Not in a bad way, but being out of batch and seeing so many new people is great, but also weird. Life moves on, people never-graduate and may not be around in the same way as before. The context change.

RC is a context. Each batch is its own context with different people coming from different backgrounds with different personalities and tastes. Each batch is big enough that I would say the average stays reasonably similar from batch to batch, but to be clear, RC is not about averages, it's about the individual connections you make and the context you bring to your batch. From my own point of view, RC was a perfect thing to do while I wait for my employment authorization to come through. But really, it has shown to be so much more than that. RC is and was a social context for me in New York. Somewhere to go and hang out, interact with others and learn new things in an open and warm environment.

RC is remarkable. I think it attracts a certain kind of people wanting to become better programmers and learn new things for the joy of it, rather than pure career progression (although one may lead to the other, the other way around maybe not so much). The directives are pretty clear: work at the edge of your abilities, build your volitional muscles and learn generously. But as much as RC asks you to apply these self-directives, I think there is a feedback loop the other way around, where the context allow you to apply these successfully.

RC as a place and context has allowed me to learn many things, make many friends and understand better what I want from a social technology context.

My focus shifted wildly during my batch. For my application I wrote that I wanted to implement a machine learning library using scheme². I quickly let go of this idea and proceeded to socialize and get in the groove of the RC spirit. For a large part of the first half of my batch I did a lot of coffee talks, learned about HTML and CSS and how to use org-mode and emacs to create this blog. A personal reflection is that I can be pretty harsh on myself. Looking back at my previous RC notes (RC-week-1.html, RC-week-2.html, RC-week-3.html, RC-week-4.html and RC-halfbatch.html) I summarize them below.

First off, I thought that my web-development learning would be swift and I would go on to do "bigger" things. In reality, this part took longer, on the other hand, I actually built this website by hand (a labour of love!) and in itself that is something beautiful. My main project (which did not materialize at the end during the batch, but I have some vague sense that this will be finished in the future) started forming in my mind early and I still have the data around. I did not finish it. I did do several presentations and non-presentations, but could maybe have done better. However, I did still present! Looking at the presentations have really been inspiring to me, enforcing the rule of learning in the open and sharing what we've learned.

For impossible day I learned to how to use flask to set up a web-server which was great. Got into the weeds on how to use databases, minimal html and elements used for functionality through flask. One more step to becoming a full-stack ML developer (I jest.. or?). This was a step forward, but I still felt a bit unfocused. As a side-project I learned some algorithms and data structures, and I can say that this has actually been a success and will hopefully help me land a job soon.

A thing I did not go into much was that I levelled up my toolsets and personal workflow. I will probably make a note about this later but in short

• Fixed a satisfactory org-mode + jupyter kernel workflow which actually works.
• Set up a VPN so I can connect to devices on my home network, which allowed for the above computational notebook to actually be possible in the first place.
• Bought a GPU and installed it.

All in all a great outcome, and possibly in the spirit of RC?

I also learned Jax and Diffusion models, together with some mechanistic interpretability, mostly for the family of LLMs and transformers. This was lead by the great Changlin Li. Dipping into ML in this way has been great and re-ignited a passion again.

So where does this leave us (me)? Three months seem long, but it's short. Chipping away day-by-day is the only way to keep going though and does lead to returns, essentially yielding cumulative interest on what you have learned and you knowledge. I feel like a more apt and capable technologist and understand how computers work better and how to use the web to share more openly. This was one of my goals coming into RC so this has been a great success. And this by itself made RC worth it.

I will stick around, at least for the foreseeable future. I will also hang around on the Zulip, so if you want to reach me, either email me (you can find my contact in the footer) or reach out on Zulip. I promise I won't bite 😄.