VC.

PERSONAL PORTFOLIO / N1 AI SCHOOL

Vivian 曹雪维.

From the a priori to the a posteriori.
From logic to experiment.

I used to look for answers within the system. Now I also want to see what happens when I put it to the test.

Explore my work
Enter another layer ↗

01 / PROJECTS

Projects

Gridline Arcade

A live two-player arcade for Connect Four, Ultimate Tic-Tac-Toe, and strategy variants.

Open the arcade ↗

Modal Logic Lab

An interactive modal logic tool. Build possible-world models, change which worlds are accessible, and use countermodels to test validity across logical systems.

Open Logic Lab ↗

Axiia Cup

An AI agent strategy tournament built by Paideia. Players write prompts for opposing roles, run matches, and review battle reports to understand strategy and adjudication.

Live product ↗ · Source on GitHub ↗

My contribution

I designed scenario-based multiple-choice questions, worked on the Prompt Builder and the first-match experience, and contributed to frontend design and debugging.

I interviewed users, designed a human testing system, and participated in review syncs and chain-of-thought reviews. I also designed and ran benchmarks to test AI judge bias and how sensitive rulings were to changes in the inputs.

02 / WRITING & FOUNDATIONS

Writing & Foundations

Philosophy & writing

My interests include philosophy of mind, language, and mathematics. I also serve on the North China organizing committee of the Secondary School Philosophy Conference. Below are three Foundations preprints and one of my competition essays.

Philosophy of mind

The Unobjectifiable

Examines whether first-person experience can be treated as a variable in third-person scientific explanations.

Read preprint ↗

Philosophy of language & science

Two Layers of Vagueness

Uses the heap paradox to distinguish our tolerance for small changes from the question of what makes “heap” true.

Read preprint ↗

Logic & philosophy of mathematics

The Philosophical Limits of Gödel’s Incompleteness Theorems

Distinguishes what the theorems establish from the extra premises needed to draw conclusions about minds, mathematics, or knowledge.

Read preprint ↗

2026 John Locke Essay Competition · Philosophy Commendation

What consolations does philosophy offer?

My essay distinguishes consolation through an answer from the relief that comes when we question why an answer seemed necessary.

Read essay (PDF) ↗

Foundations

Editor-in-Chief

I developed the journal’s original proposal and defined its scope around foundational questions in logic, mathematics, science, and AI. As Editor-in-Chief, I helped shape an editorial process built around feedback and revision, including an optional AI collaboration appendix where authors can explain how they used AI and which judgments they made themselves.

03 / NOW

mini-GPT

An experiment changed what I count as progress

I used to want an explanation for why a method should work before trying it. While training mini-GPT, I watched validation loss fall even as the model still contradicted itself in its continuations. I wondered about the data first. What was I measuring, and what did I expect that number to tell me?

After short-answer fine-tuning, the model got some copying and extraction tasks right. It also started all eight yes/no reasoning answers with “Yes” and missed all four whose correct answer was “No.” I hadn’t anticipated that failure. Now I’m more willing to test an intuition with a small experiment, then let the result change the question I ask and the explanation I give.

At N1, I started by writing a BPE tokenizer and running controlled comparisons on TinyStories. I went on to pretrain a 408M-parameter model on 6B tokens of FineWeb-Edu and compare fine-tuning recipes on a fixed set of prompts.

View source ↗ · Read the experiment report ↗

Earlier experiment: RoPE

I compared fixed frequencies and learnable rotary position embeddings on TinyStories using a model with about 14.56M parameters. The confirmation runs used three random seeds per method.

Mean held-out test negative log-likelihood (NLL, lower is better)
MethodMean NLL
Fixed RoPE1.042593
Shared learnable log frequencies1.042084
Randomly initialized learnable frequencies per layer1.041524

The per-layer variant averaged about 0.10% lower NLL than the fixed baseline and won on two of three seeds. The difference is small and comes from one dataset and model size.

In an exploratory four-choice, single-fact retrieval task with a 512-token gap, the random per-layer variant averaged 86.98% accuracy versus 55.73% for fixed RoPE. At a 2,048-token gap, all variants were near chance.

Held-out test NLL (lower is better)
Fixed 1.042593, shared 1.042084, random per-layer 1.041524
Single-fact retrieval accuracy (%) by token gap
At a 512-token gap, accuracy is 55.73%, 83.85%, and 86.98% for fixed, shared, and random per-layer RoPE. At 2,048 tokens all are near 25%.

Fixed RoPEShared learnable log frequenciesRandom per-layer

04 / RECOGNITION

Competitions & awards

  • International Logic Olympiad (ILO)
    2026 Senior Division · 4th in China · Team China · National Gold · Finalist Award · Qualified for the global final round at Stanford
  • International Philosophy Olympiad
    2026 China national team selection · First Prize (one of six nationally)
  • John Locke Essay Competition · Philosophy
    2025 High Commendation · 2026 Commendation
  • 2025 NEC Global Round (Asia) · AS Division
    Overall Team Award — Gold · Individual qualifying test — Top 10
    See all certificate distinctions
    • Team Top 10 in each qualifying test: Microeconomics, Macroeconomics, and International Economics and Current Events
    • Team Top 10 in Critical Thinking
    • U20 Youth Voice Premier Distinction Award — Silver
  • University of Pennsylvania Liberty Bell Classic
    February 2025 · Informative Speaking Finalist
  • Middle School Tournament of Champions (TOC)
    2025 · Informative Speaking National Champion

05 / EDUCATION

Education

Grade 8
BASIS International School Chengdu
Grade 9
South Hills Academy
Grade 10 · Current
Gap year, studying at N1 AI School

06 / CONTACT

Contact

WeChat
caoxuewei0330
Email
viviancao201103@hotmail.com