Hank Qiu

In the summer of 10th grade, I was in the Xplorer Science Research Program, hosted by Tencent in the CS group. That's when I realized:

1. I don't hate research and AI machine learning is more accessible than I thought — even to ordinary high schoolers like me.

2. It’s interesting to fine-tune models and try new ways to use existing literature and methods to come up with our own to beat the existing literature.

I interned at NNKosmos for two summers. The first summer, I assisted the product manager and created user experience maps and flowcharts for various products the company offered. The first product was a big vertical display screen with a digital avatar that interacts and speaks to you. Another product was a near real-time generation of digital humans for short videos and live streams — you would feed in input for what you wanted the avatar to say, and it would generate speech and animation accordingly. I mapped out how these products would be used by customers and how to improve them.

The second summer, I worked as an assistant to the R&D manager. It was much more technical since the product was already mature. I did a voice cloning project where if I have your voice for 10 minutes, I can use that as training data, post-train the model, and input any words to be spoken in your voice. Theoretically, with just 10 minutes of high-quality audio, you can make someone say anything you want – which is a major AI security and governance risk. [Laughs]

It sounds at least 80% real, and if you're not familiar with someone’s voice, it’s hard to tell the difference. That’s why you should be careful about giving your audio files to others. Once someone has enough pictures and videos of you, they can make any video they want.

This is one of the parts where it gets more questionable and debatable: whether AI is good or bad.

There's a lot of positive aspects that are brought by AI for our generation. Although AI reliance is a thing, a minority of people are harmed by it. Proper governance or AI safety is necessary, but I think there's a good share of users that are benefiting from it and you can see the productivity boost quite clearly with how quickly new apps and new elements are being developed.

Even if they're vibe-coded, if it works as a prototype, there must be some value right? If you can automate that process, why not?

At a minimal level, you need consent to carry an AI of anyone. With deepfaking, it's getting to the point where it's hard to tell the difference – even for experts in the field. Some of the mainstream models, for example, Google's Nano Banana and Veo models have SynthID watermarking in every single piece of media that they generate. With this, you’ll always know if it was completely, partially, or not created by AI. But you can also always run open source models that don't have these watermarks.

It's hard to control. For people who are so easily misled by fake news, this would just be a nightmare situation [Laughs]. Our society rewards us without thinking about the consequences.

I've always been fascinated by the idea of a digital twin simulation for economics. I think there’s potential to use AI to simulate various real-world agents, have them interact, and then draw economic conclusions from those simulations. Especially for macroeconomics, it could be a good way to come up with new theories and quickly test hypotheses.

One of the projects I'm working on right now is for the Economic Policy Institute at the UChicago chapter of the Charles River Economic Labs. We’re creating a simulation to model the impact of implementing paid family and medical leave in the US, especially in the southern states, because the US is one of the only OECD countries without it [Laughs]. Some states have already implemented it, but most still have not.

It’s important to improve accessibility of technology, especially for those with no internet or computer access. Distributing technology more widely could make an impact on fighting poverty and spreading opportunity. In the Aspiring Xplorers Science Program at Tencent, we did charity work using an AI model to detect hearing impairments. We went into the community, and people came to test whether they had hearing issues or not.

There’s little awareness of hearing disability and limited access to this kind of technology, yet people need it. The earlier you find out you have hearing impairments, the sooner you can get a hearing aid and improve your life. Sometimes you don’t even realize you need it. The more people you give useful technology to, the more it can help a community. It makes a really big difference.