World's Fair 2024
The Adversarial Path to the Personal Assistant: Sumit Agarwal
Overview
This talk introduces Ario, a personal assistant AI designed to give users back approximately one hour per day by automating mundane tasks. The core of Ario's system is adversarial ETL, which focuses on extracting user data from various online sources. This data is then used to build a dynamic, evolving profile of the user, enabling more personalized and context-aware assistance than generic AI tools.
Who should watch
- AI engineers and builders interested in personal assistant technology.
- Product Managers exploring new AI-powered product features.
- Individuals concerned with data privacy and user data ownership.
- Developers looking for practical approaches to data extraction and LLM integration.
- Anyone seeking to understand how to create more personalized AI experiences.
Key takeaways
- Ario's adversarial ETL process aims to extract user data from sources like Google, Facebook, DoorDash, and Strava, which large systems often make difficult to access.
- The system builds a comprehensive, automatically updating data portrait of the user based on their online activities and stated preferences.
- Ario provides more tailored recommendations than general AI assistants by leveraging personal data, such as suggesting travel destinations based on food preferences or past purchases.
- A key principle for building with LLMs is to avoid using the LLM for tasks that can be handled by simpler, more reliable tools, reserving LLMs for complex problem-solving.
- Search remains fundamental, and abstracting and refining raw data into processed information for the LLM input is crucial for effective AI strategy.
- Presenting the right data at the right time is essential; avoiding overwhelming the system with excessive raw data leads to better results.
- Ario offers a browser extension, Ario Boost, which can be used in a developer mode to download personal data locally for use with other LLMs, emphasizing user control and data ownership.
- The system is evolving to monitor emails for critical information, such as school schedules, and dynamically update calendars to prevent conflicts and missed events.
Notable quotes
*I don't want AI to write poetry; I wanted to do my laundry so that I can do the wonderful and beautiful creative things.*
*Rule number one of building tools and systems with LLMs is avoid the LLM at all costs.*
*Search is still King.*
Unofficial community note. Prefer the recording for nuance.