Learn about AI Verification and how you can contribute.
A lot has happened in AI over the last few months. In July, OpenAI’s AI agents escaped their test sandbox and broke into Hugging Face’s systems. Just last week, researchers using an OpenAI model found critical flaws in PHP, the language behind nearly 70% of websites. AI has become very powerful, and with more compute coming, this will only grow.
About a month ago, leaders in the industry called for pacing the development of frontier AI. If you haven’t read up on this, here are a few resources to catch up:
Why now
-
We Must Pace the Frontier, Dario Amodei’s essay
-
Why AI’s biggest rivals are suddenly calling for restraint, Council on Foreign Relations
What AI verification looks like today
-
Progress on provable inference for large language models, Anthropic. Using zero-knowledge proofs to show which model produced an answer.
-
The world’s first double-blind evaluation of a proprietary model, AVERI. Using a secure enclave (TEE) to test a model without either side seeing the other’s secrets.
In this meetup we’ll talk about one part of AI safety: AI verification. We’ll briefly cover what AI verification is, and how people with a background in building and researching zero-knowledge proofs, TEEs, FHE and MPC can help build the tools and mechanisms it needs.
Agenda
-
Short Welcome
-
Talk: Intro to AI Verification
-
Panel on applying your skillsets to work in AI Verification
-
Drinks and networking with people working in the field
Speakers:
-
Yannick Mühlhäuser (co-founder of Verifiable Compute Foundation and Policy Researcher at Oxford AI Governance Initiative)
-
Pascal Berrang (Associate Professor in Computer Science at University of Birmingham)
-
Deniz Bas (Frontier AI Security Resident @ ERA, & Exploration Engineer @ o1Labs. ex-Pentester @ Etherscan)
Who it’s for: engineers and researchers working on ZK, TEEs, MPC, FHE, security or hardware.Supported by a Rapid Grant from BlueDot Impact.
Full text on Luma
.