Rachel Lim is an Engineer on the inference team at OpenAI, where she works on the large-scale systems that enable AI tools to run efficiently and reliably. A Singaporean, Rachel is based at the company’s headquarters in San Francisco. OpenAI’s Asia Pacific hub in Singapore works with local partners and businesses to expand access to AI and support its growing adoption across industries, reflecting the region’s increasing role in shaping how AI is used globally.

1. Explain to us what you do within OpenAI and its inference team.

As an engineer in OpenAI’s inference team, I primarily work on AI systems. The inference team makes it possible for people to use the models and capabilities developed by research teams at OpenAI by making them more efficient and reliable over time. For example, each time you make a request on ChatGPT, Codex (our agentic coding tool), or our application programming interface (API), there are billions of mathematical operations running to generate every single token that is produced. Our job is to make sure that runs as efficiently as possible on the computers we have.

Both research and deployment involve running a large number of computers in a coordinated way, and oftentimes, doing things well requires understanding the machine learning (ML) operations happening under the hood. It sounds complicated, but a lot of the principles underlying this are simple and are things that come up in our day-to-day life. For instance, it’s like going with your family to the hawker centre and splitting up so that you can order three dishes from three different stalls at the same time — this is parallelism. Or sometimes, if you’re at a restaurant and are in a rush, you’ll ask for the bill before you’re done eating because you know it’ll take the staff time to prepare it, so you can leave as soon as you’re done eating— this is pipelining. Running all of the computation behind a model is similar, except the parallelism and pipelining are done across the hardware that we use. In a typical day at work, I often look at dashboards to understand how our systems are doing — whether they’re performing as expected or if things are slow because of errors — thinking about problems and how to solve them, discussing ideas or project milestones with collaborators, and reading and writing (or even deleting!) code. Thankfully these days, Codex writes a large amount of code for me.

2. You’ve worked with AI in other tech companies before joining OpenAI. Why are you passionate about tech, and what’s it been like for you at OpenAI?

I first learned to code in university. It was somewhat serendipitous actually: my roommate in my freshman year was a computer science major, and she inspired me to take an introductory programming class. I took to it almost immediately, partly because the class allowed me to build things that I thought were cool (like a brick breaker game!) within the span of a few weeks, and partly because a lot of the day-to-day work of programming felt like solving puzzles with logic and reasoning, which is something I’ve always enjoyed.

From the outset, I was also drawn to the idea that technology could be used to solve problems that were meaningful to people, or that hadn’t been solved before. When I graduated from university, deep learning was already starting to take off though it was in early stages compared to now and had much narrower use cases than what we see today. Naturally, I was drawn to working in ML.

I joined OpenAI in 2020, right after the company published its GPT-3 paper and released the model through the API. I had heard a research presentation about it and found it compelling: deep learning had been gaining traction for a while, but here was something that seemed to hold promise for being useful in a more general way, and scaling laws suggested it would just keep getting better. This, in fact, has panned out. For me, it's been really fun not just to get a front row seat to the action, but to be able to contribute towards building products and capabilities that have never existed before – things people find useful, and that only existed in the realm of science fiction a few years ago.

We’re trying to build artificial general intelligence (AGI) which can benefit all of humanity. This means doing research that pushes the frontier of what AI can do, and deploying it in a way that is accessible and useful to humanity writ large. This can take many forms, whether it’s an assistant that helps you with your day-to-day tasks, or a tool that speeds up scientific discoveries which can then improve the lives of people.

Within the region specifically, we have offices in Japan, Singapore, India, South Korea, and Australia. Asia-Pacific is one of our fastest growing regions for ChatGPT, and we see strong adoption around Codex, our agentic coding tool, as well. This is why we’re hiring a lot regionally, especially in Singapore. The technology is already capable, but there’s still a lot more work to do to bridge things so that people find it useful where they are. We’re always looking out for people who can help do this, and partnering with local enterprises, developers, universities, and public services to make our technology more accessible and useful.

3. There are many anxieties around how AI is being used in our society. As someone in the industry, what are your views on this?

I think AI, like many tools and technologies before, holds both a lot of promise and peril. When used well, it empowers people, provides them with greater access to helpful information, and solves problems that haven’t been solved before. But when misused, you can get misinformation or AI slop (think low quality and bizarre AI-generated content). There are also more systemic risks involved like economic displacement or the undermining of existing systems.

It’s complicated, and frankly it isn’t easy to strike the right balance between empowering the positive use of AI while also trying to prevent bad outcomes. We still have a lot to figure out within the industry, even within OpenAI itself, but one thing I find sensible is how companies like OpenAI are deploying the technology gradually and iteratively. This means people can reap the benefits of AI while learning and getting better at mitigating the potential harms along the way. I think it’s also helpful to bring people along in a transparent way as the technology develops so that we, as a society, can figure out what to do and be ready for what’s coming next.

4. How do you think people can future-proof their careers in a world that’s increasingly adopting and embracing AI?

I think the most important thing is to be adaptable. We live in a chaotic and ambiguous world, even more so with AI. The shape of my job now is so different from when I first started, not just because I’ve grown in experience, but also because we now have these coding agents that have transformed the way we work. I still love my handwritten artisanal code, but working in such a fast-moving field, it’s important to be versatile and to not be too attached to any particular facet of your job. Most of the time, we have to just roll with it. Believe in yourself and your resilience.

5. Why do you think AI is an exciting field to be in, and what would you do if you were a fresh graduate interested in exploring the power of AI?

There’s never been an easier time to be a builder than now, especially for those without a tech background. Learning to program really changed how I saw myself in the world. It’s really empowering, and I think everyone should learn to do it, whether it’s writing code yourself or vibe-coding. It’s always felt magical to me that you can want something to exist, and then create it yourself.

If I were a fresh graduate in Singapore, I would spend more time daydreaming about the applications that AI can enable, and work to create the ones I’m most excited about, because if you can think it, you can definitely build it one way or another.