informedgirl.com
Up-to-date analysis of the latest findings on women's health, skincare, anti-aging, pesticides, and microplastics.
I like to take systems apart and ask why they were built that way.
scroll ↓
✦
I grew up in the Bay Area, lived in China on a Fulbright, and now call Zürich home. Along the way I've worked at Google, co-founded a YC and Techstars-backed startup, and spent the last few years in crypto thinking about how decentralized systems actually distribute — or fail to distribute — power.
These days I'm interested in AI, what it means for geopolitics, and whether the people building it are asking the right questions.
I think in English, grew up in Russian, learned Mandarin just because it was hard, gave my heart to Spanish, and am currently losing a battle with German grammar.
Main quests
Now
Head of Product Design, Offchain Labs (Arbitrum)
If it has pixels, I run it: arbitrum.io, portal.arbitrum.io, bridge.arbitrum.io, and offchainlabs.com. Design for an ecosystem securing $17.6B.
Highlights
- Age 10, got my first job by posting a Craigslist ad. Age 24, founded a SaaS company, acquired by HubSpot. Forbes 30 Under 30.
- Security analyst at Google, then product designer
- Melanoma researcher at Stanford; qualified for an undergrad position while still in high school
- Researched the technical infrastructure of censorship in China under a Fulbright
Also
Product management, drone robotics
FACES, APW, SABF, BTIC - an alphabet soup of youth leadership grants in China, Australia, Argentina, and New York
Side quests
Selected work
Things built with code, curiosity, and AI.
Recent thoughts
Musings in the human condition.
Thoughts
Musings in the human condition.
Projects
Things built with code, curiosity, and AI.
ai-incest.com
Exploring how AI's are related and who might be able to audit who.
Verifiable Claude
Borrowing fraud proofs from blockchain to build a verification layer for AI. What if Claude had to prove its claims?
Recently I asked Gemini for a synopsis on a book and it gave me completely wrong information. I told it it was wrong, it apologized. Then proceeded to give me different but still totally wrong information. I told it "only tell me information you can verify is fact, do not tell me something if you don't have access to the contents of the book itself, just tell me you don't know". And it STILL confidently gave me totally wrong information.
And so I reached the jagged frontier of AI, where the magic is less sparkle and more dog poop.
In blockchain, particularly in my world of Ethereum L2s, we have this thing called fraud proofs. If someone lies about the state of the chain, there's a system of checks to verify and punish the liar.
So what about fraud proofs for AI?
LLMs hallucinate with confidence. Blockchain systems fail explicitly.
In Arbitrum (a blockchain):
- If a sequencer makes an invalid state transition → fraud proof catches it
- If data is unavailable → the system halts, doesn't guess
- If verification fails → transaction reverts, doesn't pretend it worked
In AI agents:
- If the model doesn't know → it makes nonsense up confidently
- If retrieval fails → it fills in plausible-sounding baloney
- If verification would help → there's no mechanism to verify
I decided to borrow from blockchain architecture and build fraud proofs for AI. Optimistic Rollup is a pattern where systems assume data is valid but enable verification. (Hence, "optimistic")
Verifiable Claude
When Claude makes a factual claim, it should provide a fraud proof — evidence that can be independently verified.
- Optimistic execution: Claude responds quickly with claims
- Challenge period: Users can flag suspicious claims
- Fraud proofs: System verifies claims against sources
Just like Arbitrum assumes validity but enables verification, Verifiable Claude assumes truthfulness but enables fraud proofs.
Tech used: Brave API for web searches · Anthropic API for Sonnet/Opus queries · Claude for vibe-coding