Zac Hatfield-Dodds – Formal Verification is Overrated [Alignment Workshop] скачать в хорошем качестве

Zac Hatfield-Dodds – Formal Verification is Overrated [Alignment Workshop]

Zac Hatfield-Dodds presents “Formal Verification is Overrated,” arguing that relying solely on verification methods may not provide real AI safety. Complex model weights exceed what current tools can handle, and simplifying real-world dynamics for verification often introduces risky assumptions. Additionally, even simple “tool AI” systems can unintentionally gain autonomous behaviors, challenging safety expectations. Highlights: 🔹 Model Complexity – Even small models with billions of parameters are beyond the capacity of current verification tools. 🔹 Uncertain Reality – Verifying AI against a world model requires assumptions that may not hold up. 🔹 Tool AI Instability – Tool-based AI might not remain tools, with even simple functions risking unsafe autonomy. The Alignment Workshop is a series of events convening top ML researchers from industry and academia, along with experts in the government and nonprofit sectors, to discuss and debate topics related to AI alignment. The goal is to enable researchers and policymakers to better understand potential risks from advanced AI, and strategies for solving them. If you are interested in attending future workshops, please fill out the following expression of interest form to get notified about future events: https://far.ai/futures-eoi Find more talks on this YouTube channel, and at https://www.alignment-workshop.com/ #AlignmentWorkshop