👋 Hi, I am Heng-Yi Wu. I live in Taipei, Taiwan. 🇹🇼
💼 Work Experience
⛓️ April 2022 - Present
I am currently working at a blockchain company, where I primarily use JavaScript and Node.js.
🤖 January 2022 - April 2022
I worked at an AI company, where I primarily use JavaScript and Node.js.
📄 February 2020 - January 2022
I worked at a resume-building company, primarily using Ruby on Rails.
💻 January 2017 - February 2020
I was employed at an online course platform.
May 2019 - February 2020
I contributed to the initiation of a new B2B product using Ruby on Rails, which was launched within six months.
May 2019 - February 2020
Following a change in leadership, I assumed the role of tech lead.
The engineers I collaborated with were highly skilled and required minimal mentoring. Consequently, my focus shifted to fostering an environment of self-learning and knowledge sharing, ensuring the preservation of this culture despite the management’s shift in focus towards profit. Their eagerness to learn often led me to discover new insights as well.
Notably, one of my colleagues joined the Cargo team for the Rust programming language, enriching my understanding of open-source contributions. I deeply appreciate these experiences.
January 2017 - May 2019
During this period, I primarily used JavaScript and Node.js in production.
I developed an iOS app using React Native; however, it was not published due to Apple’s strict payment policies.
👔 June 2016 - November 2016
I worked at a human resources platform focusing on software engineers.
Unfortunately, the company closed after half a year due to misalignment between the founders, leading me to seek another job.
In this role, I primarily used Ruby on Rails in production and met my mentor, who guided me in using Amazon Web Services and infrastructure tools like Terraform.
Hobbies
🤖 Trying new LLMs
My primary focus is on security and censorship issues.
To avoid legal complications, many Big Tech have placed extensive restrictions on content generated by numerous LLMs. One such LLM, named Goody-2, has been created to mock these constraints. In my experience, I once tried to summarize an article about a supposed sexual harassment case.
I noticed that the summary, built token by token, was ultimately replaced with an error notification denoting inappropriate content. This leads me to believe that OpenAI might be using a post-generation filtering mechanism to block access to certain summaries. I view this as excessive censorship since the article primarily dealt with the legal dimensions, rather than the actual instance of sexual harassment.
I hope that one day, LLMs will become censorship-free and truly open, meaning I can reproduce the same model using instructions provided by the maintainers, much like how we use software today.
👨💻 Exploring new programming languages
The idea of linguistic relativity, also known as the SapirāWhorf hypothesis (/sÉĖpÉŖÉr ĖhwÉĖrf/ sÉ-PEER WHORF), the Whorf hypothesis, or Whorfianism, is a principle suggesting that the structure of a language influences its speakers’ worldview or cognition, and thus individuals’ languages determine or shape their perceptions of the world. - Wikipedia
To encourage out-of-the-box thinking when building things, I enjoy experimenting with new programming languages. In addition to the notorious Haskell, I’ve also tried Elixir, a functional programming language that borrows many syntax elements from Ruby.
Recently, I’ve been dedicating most of my time to Rust, aiming to familiarize myself with memory safety. I also invest time in Nix and NixOS, striving to build a reproducible development environment for daily use.
🍓 Self-hosting with a Raspberry Pi
A journey of a thousand miles begins with a single step. - Laozi
Every piece of software begins small.
As an engineer, I’ve seen many in my field quickly dive into creating vast, intricate systems and infrastructures for minor issues. Being a “senior” engineer sometimes means I’m expected to do this too, but it’s not an approach I prefer.
Self-hosting software on a Raspberry Pi for my daily needs helps me adhere to the principle of simplicity. The Raspberry Pi’s resource constraints limit the viability of elaborate distributed systems. This pushes me to bypass complex infrastructures, concentrate on the essentials, and then carefully optimize.
I constantly remind myself that the simplest solution is often the best one. It may not be flawless, but as long as it works, that’s usually sufficient.
More often than not, these straightforward solutions prove to be effective over the long haul, even as the high levels of traffic we sometimes anticipate never actually come to pass.