Bret Taylor discusses the AI market's shift to autonomous agents and outcome-based pricing, the future of coding with AI, and strategic advice on GTM, pricing, and where to build in the new AI landscape. He shares career-defining lessons from Google, Facebook, and Salesforce.
Sriram Krishnan, Senior White House Policy Advisor on AI, outlines the America AI Action Plan, a strategy designed to ensure U.S. dominance in artificial intelligence. He discusses the three core pillars of the plan—infrastructure, innovation, and global standards—while also exploring the geopolitical race with China, the critical role of open-source models, and the need for America to own the full AI stack, from GPUs to applications.
A technical deep dive into Apple's Private Cloud Compute (PCC), exploring its novel architecture for running sensitive AI workloads with cryptographic privacy guarantees. The talk covers the core requirements, key components like remote attestation and transparency logs, and how these concepts can be applied by developers today.
A security analysis of YC AI agents reveals that the most critical vulnerabilities are not in the LLM itself, but in the surrounding infrastructure. This breakdown of a red teaming exercise, where 7 out of 16 agents were compromised, highlights three common and severe security flaws: cross-user data access (IDOR), remote code execution via insecure sandboxes, and server-side request forgery (SSRF).
OpenAI's Head of Education, Leah Belsky, and students Yabsera and Alaap discuss how AI, particularly ChatGPT and its new Study Mode, is transforming education. They cover global adoption, the shift from policing AI to integrating it, its role as a personal tutor for building confidence and skills, and the evolving nature of learning, work, and critical thinking in the age of AI.
Fouad Matin from OpenAI's Agent Robustness and Control team discusses the critical safety and security challenges of code-executing AI agents. He explores the shift from models that *can* execute code to defining what they *should* be allowed to do, presenting practical safeguards like sandboxing, network control, and human review, drawing from OpenAI's experience building Code Interpreter and the open-source Code Interpreter CLI.