Unmasking the Shadows: AI Red Teaming in the Age of Gemini and VertexAI, fortified by SAIF

The rapid advancement of AI, exemplified by models like Gemini and the capabilities of VertexAI, brings forth not only unprecedented opportunities but also a new landscape of security risks. As AI systems become increasingly integrated into critical applications, ensuring their robustness against adversarial attacks is paramount. The Secure AI Framework (SAIF) offers a structured approach to building security into AI systems from the ground up.

This talk goes into the world of AI red teaming, a proactive approach to identifying and mitigating vulnerabilities in AI systems, and how SAIF can bolster these efforts. We'll explore:

- The evolving threat landscape of AI, from data poisoning to prompt injection attacks.
- How AI red teams operate, simulating real-world adversaries to uncover weaknesses.
- The role of SAIF in proactively addressing security concerns during AI development and deployment.
- Practical strategies for incorporating AI red teaming and SAIF principles into your development lifecycle, ensuring your AI systems are resilient against emerging threats.

Join us to gain insights into the critical role of AI red teaming and SAIF in safeguarding the future of AI, particularly as we navigate the complexities of increasingly sophisticated models and platforms like Gemini and VertexAI.

Turbocharging Web Apps: Efficient AI Model Caching in Chrome

The integration of AI models into web applications is revolutionizing user experiences, but the large size of these models poses a significant challenge for performance. Slow loading times and unresponsive interactions can frustrate users and hinder adoption. This talk will delve into the critical aspect of efficient AI model caching in Chrome, exploring strategies to optimize model delivery and ensure seamless user experiences.

We'll cover:

- The performance implications of large AI models in web applications.
- Best practices for configuring cache headers to leverage browser caching effectively.
- In-depth comparison of client-side caching techniques: Cache API, Origin Private File System (OPFS), and IndexedDB.
- Advanced strategies for handling user-selected models and large file downloads.
- Security considerations and trade-offs when caching AI models locally.

Whether you're a web developer, AI engineer, or performance enthusiast, this talk will equip you with the knowledge and tools to deliver lightning-fast AI-powered web experiences. Learn how to optimize model loading, reduce latency, and create web applications that seamlessly integrate AI capabilities without compromising performance.

LLMinABox: On Device Personalized Diary And Concierge using your voice and Gemma

Supercharge your productivity and learning with LLMinABox! This talk unveils an AI-powered solution that transforms your device into a personalized diary, instant meeting assistant and concierge. Harnessing large language models (LLMs) and on-device processing, LLMinABox offers seamless voice interaction and intelligent assistance, all while safeguarding your privacy.

**Key Features:**

- **Effortless Voice Capture:** Capture thoughts, ideas, and **meeting notes** using natural language. LLMinABox transcribes and organizes your recordings, making them easily searchable and actionable.
- **AI-Powered Meeting Summaries & Action Items:** Generate concise and informative summaries of your meetings, highlighting key points and **automatically extracting actionable items**.
- **Personalized Learning Tools:** Generate tailored study materials, code examples, and explanations based on your voice recordings, meetings or uploaded notes.
- **On-Device Processing:** All data stays on your device, ensuring complete privacy and security.

**See LLMinABox in action with live demos and code examples. Learn how to leverage LLMs for on-device, privacy-preserving AI solutions.**

Give your web apps superpower with Generative AI and Mediapipe

Learn how MediaPipe empowers web developers with custom, cross-platform ML solutions for enhanced web applications. No deep learning expertise? No problem! Discover how to effortlessly create and implement generative AI solutions, unlocking unlimited potential to transform and elevate your web app.

Here are some of the topics that will be covered in this talk:

What is generative AI?
How does MediaPipe work?
What are the different types of generative AI models that MediaPipe supports?
How to choose the right model for your needs
How to create custom ML solutions with MediaPipe
Examples of custom ML solutions that have been created with MediaPipe
This talk is for web developers who want to learn how to use generative AI to supercharge their web applications. No prior experience with ML is required.

Immersive Art with Generative AI and Intelligent Captioning

In this talk, we'll explore the exciting convergence of web technologies, artificial intelligence, and artistic expression. We'll delve into the world of WebXR, showcasing how it can transform static art galleries into dynamic, interactive 3D environments accessible directly through web browsers.

We'll then dive into the transformative power of generative AI models like Vertex AI Image Generation. Learn how these models can create unique, ever-changing artwork on demand, opening up new possibilities for artistic exploration and audience engagement.

But the experience doesn't stop there. We'll demonstrate how intelligent captioning, powered by models like Gemini Multimodal API, can add depth and context to each artwork. These AI-generated captions can provide insightful descriptions, interpretations, and even historical context, enriching the viewer's understanding and appreciation.

On Device Generative AI: Building your own Dall-E in the browser, welcome WebGPU

Have you ever wanted to create your own AI-generated art? Well, now you can! In this talk, we'll show you how to build a Dall-E-like image generation app in the browser using the WebGPU API.

WebGPU is a new standard for graphics processing that is designed to run on the device itself. This means that you can create powerful and user-friendly AI applications that run right in your browser.

We'll start by giving you an overview of on-device generative AI. Then, we'll introduce the WebGPU API and show you how to use it to build your own image generation app. Finally, we'll discuss the challenges and opportunities of on-device generative AI.

So come along and learn how to create your own AI-generated art!

Rapid Prototyping using Gemma and Production deploy in Vertex AI

It's easy to get started with Gemma! In this walkthrough, we will guide you step-by-step, from finding the model, pulling up code snippets, and creating a simple prompt in just a few clicks. Navigating your favorite integrations and tools has never been so easy. And learn how to run Gemma locally on your laptop using quantized models.

Web AI: New models, tools, and APIs for your next web app

learn what's new with the state of Web AI at Google and discover Gen AI models you can run client-side in the browser, along with new tooling and APIs for your company's next web app. I'll cover key web ML updates from Google Chrome, TensorFlow.js, MediaPipe Web, Core ML, Visual Blocks, and beyond. Suitable for everyone.

Practical on-device AI for web developers

With new capabilities landing in the browser, it's becoming possible to run AI and machine learning workloads on the client side – which can drive latency, cost, and privacy wins. We'll take you through real-world use cases and discuss when on-device AI is the right choice. We'll show how new web capabilities in Wasm and WebGPU are enabling even more use cases. Finally, we’ll talk about remaining challenges, and share some early thinking to kickstart a collaborative effort.