Keynote presentation

Custom copilots are evolving from simple LLM+data retrieval augmented generation (RAG) solutions into more complex solutions with multiple interoperating APIs, models, metaprompts and grounding data. In this session, we’ll talk about these emerging patterns and show how Azure AI Studio empowers developers to build multi-agent copilot solutions. Discover advanced features such as debugging tools and security measures for developing intelligent, scalable, and safe AI solutions.

The talk will start introducing Roblox and the content creation journey of creators who build experience on the Roblox platform. AI is fundamentally changing the process of creation and is disrupting content creation workflows that need to be re-imagine from the ground up. The talk will go through some of the approaches, key findings and strategic bets in the Generative AI space, as well as future challenges ahead.

While in today’s AI landscape bigger seems to always be better, there is a burgeoning field on the other end of the spectrum. This talk will cover the basics of on-device AI model development and product applications of this technology at Instagram.

In AI development, moving from hype to high-value means optimizing base models into specialized versions. This presentation shows the benefits of smaller, focused models that outperform proprietary mega-models ones in latency, throughput, and cost-efficiency. Learn practical methods, real-world examples, and best practices for moving from prototype to production. Discover how Fireworks AI helps developers go from idea to application with the fastest and most efficient inference platform. Additionally, explore how AI developers are creating compound AI systems, combining multiple models and components to tackle the most challenging AI tasks.

Facebook's new tool, Write with AI, is designed to enhance user-generated content by helping users overcome common writing challenges like writer's block and tone misalignment. Leveraging the advanced LLaMa3 model, the tool offers real-time editing suggestions and personalized responses to user prompts, significantly improving the clarity and engagement of social media posts. Since its launch, Write with AI has increased the overall share rate of posts, indicating a substantial improvement in user satisfaction. The tool is set to expand further, incorporating features like auto-completion and contextually rich photo captions (LlaMa3-multimodal), aiming to transform digital communication on Facebook by making it more personal and expressive.

In recent years it’s become much easier to talk to machines freely. But rather than talking to machines ourselves, can we have machines talk to others for us? Communication (e.g. phone calls) is an API to the real world. By teaching machines this “protocol of human conversation” we can tap into any service built for humans with instant deployment at zero integration work by any remote party, creating value for both sides. Of course, doing this at scale in consistently high quality and guaranteeing safety and privacy remains challenging. In this talk we bring several tips and advice from having deployed the technology called Google Duplex to several such applications in production, at scale.

First will speak about the history of generative AI Media products in the industry, followed by products at Meta itself. Then the product development journey / learnings from it, followed by lessons from the launches themselves. Talk about new stuff that will be launching before the talk but I can't write here yet. Talk about the future and what's to come and what challenges are we excited to take on for those future products.