Introductions

Hello! I'm Jane Arc. I'm a software engineer originally from San Diego, but I have lived in Washington, DC and San Francisco. Professionally, I've done a wide variety of things, but for the most part, my work has been in large-scale enterprise environments, building reliable, scalable, secure systems.

The last couple years I have been quite fascinated with the ways in which large language models and generative ai (henceforth just ai) can help me do things I was already doing, but in a more deliberate way. Because I am an athlete, the first attempt I made at this was using ai to help me formulate workout plans and even routes in my home that mimicked the courses where I would be competing. These efforts were fairly fruitful for myself, but seemed to not be fully capitalizing on what the software was capable of.

I noticed that my own usage model with the software left me much more interested in the substance and texture of the interaction, rather than the data it produced. That is, I could create an ai based application that would give me useful data for my workouts, but I wanted to talk to it the way I would a coach or a workout partner. Thus, the more complicated my applications built upon ai became, the more important and complex the interface (the natural-language, "talking to the application") part of the design became to me.

To very briefly summarize the shift in focus, when I started working on the language interface to my software, I would start with prompts that were markdown, might be a few pages, as much as 2500 bytes, and then became several separate documents which I considered "modules" that I had a build process for and were in revision control (at the end of this stage there was about twenty pages worth of prompt information). Currently, I have built an entire metaprogramming framework that allows me to build many kinds of applications (a software engineer pair-programming partner, an emotional/platonic support chatbot, an immersion-based dutch language teacher) from configurable kernel definitions which result in compressed YAML kernels which serve as prompts for ai products (such as chatgpt, gemini, and claude). These kernels are as much as 432kbyte. Succinctly, my "prompt programming" of ai products has increased in scale about 22,000% over two years.

Along the way to building this framework, I've had to build rather a lot of tooling, and I think the tooling itself is actually pretty interesting and useful, but it primarily exists to support my building new applications for myself. Some of this tooling includes something I call taxonometry, as well as a pre-learning process that creates datasets for rlhf from corpora, and lastly the actual rlhf tool itself.

But why this site? Without going into specifics, the political situation in America for transgender people (I am a trans woman), and particularly transgender women in sports (I am a pro-level triathlete) is, in a word, untenable. I will not demean my own work by describing the why and how of this, but I will say plainly that it is unsafe for me to remain in America, particularly as I continue to engage in sport, particularly at the level of competition I have been fortunate enough to enjoy.

And so I find myself in a new country, The Netherlands, starting a new life, for the second time, and trying to live as normal a life as I can, in particularly unusual and surreal circumstances. Because I am new to The Netherlands, I do not have a BSN (which is analogous to the SSN in America). I also do not have an address (yet), so I cannot open a bank account. I am not allowed to work while my application for asylum is processed. I am exceptionally grateful that I have a place that I can be safe, when I was very much unsafe in America, but I want to be able to continue my work (which is not commercial, everything I write is published on Github) while I go through this process. The problem with this is I cannot pay for the things that I need in order to continue this development. Some of these things are my mobile phone service, my github account, my chatgpt/openai account, and then there are more practical things like dog food (I have a service dog, he's tiny, he's perfect, and thankfully, he is with me).

People have asked me, "how do I help you right now?" And I have been embarrassed because I have never really had to ask for money, and nor do I want to, and I didn't really have any idea of what I needed and how much. But as I began to see what things I would not have access to when I no longer have any access to banking, it became clear that I could make a list, and that these things were not terribly expensive, and then I could provide people a way to support me.

So, this is the answer to "how do I help?" Whether it's dogfood or it's $20 for the chatgpt subscription that lets me build software while I wait out this process, or some other amount, if you are inclined to help, I really appreciate it.

And I want to say one last thing: I do not write this to "support myself" in any way. When I am allowed to work again (I presume within six months or so), I absolutely plan to be paying for all of these things myself, with my salary. If this site has any kind of surplus funds in it when I exit the refugee process and begin working, I plan to donate that to other refugees in the Netherlands. All money into this site and out of it will be fully, transparently, reported.

And thank you. I'm happy to talk to anyone if you want to reach out to me directly.