A claim, then the work that tests it
Technical progress is not automatically human progress. The site starts from that idea, then shows the systems, papers, and teaching I have used to make it concrete.
The claim
The work on this site is not a list of venues first. It is a sequence of ideas about access, compute, and judgment — and then the papers, code, and teaching used to pressure-test them.
01
If a system can generate language, images, or video, the interesting question is not only whether the output looks good. It is who can enter the conversation at all. Sign-language generation is one place where that question becomes unavoidable.
Tested in Text2Sign
02
A method that only works on a cluster is a different object from a method that can run on one GPU or on a device. Efficiency is not a performance footnote. It is part of whether a system can be used, inspected, and shared.
Tested in on-device diffusion and University Ranking
03
Fine-tuning can raise a score while quietly changing what a model attends to, forgets, or hides from later review. If people are going to rely on the result, the change itself has to remain measurable — and someone has to remain able to disagree with it.
Tested in the CLIP attention study and the notes on human judgment
A single-GPU diffusion baseline for short sign-language clips from text, with public training code and a released checkpoint.
From visibilityAn 80-run comparison of full fine-tuning and LoRA that holds the learning-rate grid fixed and tracks both attention change and zero-shot retention.
From limitsA systems implementation of fused low-bit kernels and cache-update fusion, measured on hardware rather than only as operation counts.
From limitsA bilingual search app and API that puts several public ranking sources into one catalog a person can actually query.
A closer look at how AI systems can formalize unequal treatment, with the health-care risk algorithm study as a concrete warning.
The labor issue is not only whether jobs disappear, but how AI changes bargaining power, task control, and whose work becomes more precarious.
AI is often described as pure software, but its growth depends on data centres, power systems, cooling, and public infrastructure.
The third NFLS AI Club meeting featured an invited guest lecture on how computer vision, multimodal understanding, and large language models are deployed in real industrial …
The second meeting was a 35-minute technical workshop tracing a single idea from visual intuition through rigorous mathematics to working PyTorch code: how neural networks solve …
The first formal meeting brought new and returning members together for an ice-breaking session designed to surface motivations, share curiosity, and establish the club as a real …
The NFLS AI Club officially launched at the annual school club fair, opening recruitment and introducing AI to students across the school.
A residential engineering leadership program at Georgia Tech focused on design thinking, robotics, communication, and collaborative problem solving.
Continue the argument
I am glad to talk about a claim that should be sharper, a result that does not yet support it, or a setting where the work should be tested next.
🚀Student researcher · President, NFLS AI Club
Nanjing Foreign Language School
Artificial Minds, Human Values
I’m Ruize Xia, a student at Nanjing Foreign Language School in Nanjing, Jiangsu, China. ORCID: 0009-0000-0501-0943.
The site is organized around a claim, not a venue list: technical progress is not automatically human progress. I use that idea to decide what to build next — accessible generation, systems that still work under real compute limits, and evaluation that stays visible to the people affected.
The public work is how the claim gets tested. Text2Sign asks whether sign-language video can be generated under a single-GPU budget. The CLIP attention study asks what fine-tuning changes besides the score. The on-device diffusion kernels ask whether an efficiency story survives hardware measurement. Code and checkpoints are on GitHub and Hugging Face.
The NFLS AI Club is the same argument taught in public: how a model works, then whether it should work that way, and for whom.
High School
Nanjing Foreign Language School
