Podcast Summary
Intuit's AI investment and expansion: Intuit, a leading tech company, invests heavily in AI, with 810 million annual customer interactions, 65 billion daily machine learning predictions, and 2 billion models run in production. They use AI for tasks like document understanding, transaction categorization, and personalized models, and are now expanding with generative AI through GenOS.
Artificial Intelligence (AI) has become a foundational technology for Intuit, a leading technology company known for products like TurboTax, Credit Karma, and QuickBooks. Intuit's mission is to power prosperity around the world by solving financial challenges for its 100 million customers. AI plays a crucial role in this mission by automating tasks, saving time, and improving customer experiences. Intuit has invested heavily in AI, with 810 million AI-driven customer interactions per year, 65 billion machine learning predictions daily, and 2 billion models run in production daily. They have also filed approximately 900 AI-related patents. The company has a long history of using AI for tasks like document understanding, transaction categorization, and personalized models. Now, Intuit is expanding its AI capabilities with generative AI, which allows apps to generate new content, and is being developed through a platform called GenOS. Shubhang, the chief architect for Intuit Mailchimp, and Meryn Kurian, the principal engineer and AI platform architect at Intuit, are leading this effort. AI is not just a cost-cutting measure for Intuit, but a key component of their product strategy, enhancing workflows and providing valuable insights for their customers.
AI in Finance and Marketing: AI is revolutionizing business operations in finance and marketing by automating tasks, offering personalized suggestions, and generating creative content. The democratization of AI through no-code solutions and generative operating systems is making it more accessible to everyone, accelerating its adoption and impact.
AI is increasingly being integrated into various aspects of business operations, particularly in the areas of finance and marketing. For finance, AI is used to automate form filling and financial modeling, offering personalized suggestions based on individual user behavior. In the marketing domain, AI is used for send time optimization, product recommendations, and even creative content generation. Intuit Mailchimp, for instance, has made strategic acquisitions in the AI space to provide these capabilities to their customers. From a broader perspective, there's a push to democratize AI, making it accessible to non-data scientists and analysts through no-code solutions. The latest in this series of investments is GenOS, a generative operating system designed to make AI more accessible to everyone. GenOS allows users to upload their data and generate models without requiring a feature engineering pipeline, which is typically required for classical AI. This is an exciting development as it further expands the reach and application of AI in various industries. In summary, AI is transforming the way businesses operate, particularly in finance and marketing, by automating tasks, offering personalized suggestions, and even generating creative content. The democratization of AI through no-code solutions and generative operating systems is making it more accessible to everyone, accelerating its adoption and impact.
GenOS development challenges: GenOS is a unified platform for developing Gen AI applications, addressing challenges of velocity, domain knowledge, and responsible AI, built on a combination of existing features and custom-built models.
GenOS is an operating system designed to streamline the development of Gen AI applications by providing a unified path and institutionalizing the combined knowledge of an organization. It aims to address the challenges of velocity, the need for domain knowledge in off-the-shelf LLMs, and responsible AI and data governance. GenOS functions like a Lego table, allowing users to choose and reuse available features while also enabling the creation of domain-specific solutions. The system is built on a combination of existing features and the ability for teams to contribute their own. The base of any generative AI is the large language model, and GenOS uses both existing and custom-built models. The goal is to make Gen AI development accessible and extensible for various use cases while ensuring adherence to responsible AI and data governance principles.
Domain-specific language models: Companies may need to fine-tune off-the-shelf language models with their domain-specific data to provide best practices and personalization.
While off-the-shelf language models (LLMs) can provide general knowledge about a domain, they may not be sufficient for businesses with specific needs and requirements. Companies may have invested in domain-specific language models and data to address cost, latency, and accuracy issues. The use of both custom and commercial language models depends on the situation. The knowledge and features businesses want to provide to their customers can be categorized into three buckets: general domain knowledge, best practices, and personalization. To provide best practices and personalization, companies need to fine-tune LLMs with their domain-specific data. The Genoese system, for instance, includes a layer for interactions and rendering (Gen UX), a runtime for application execution (Gen Runtime), LLMs, and a pluggable system for domain-specific capabilities. The development and management of such a system involve technology, people, and processes, with leadership buy-in and good processes being crucial for success.
Intuit's success factors: Intuit's success in building Genoese was due to a strong organizational culture, effective communication and collaboration, clear decision-making, and foundational capabilities already in place.
Intuit's success in building Genoese, their AI-driven personal finance assistant, was due to a combination of factors. These include a strong organizational culture, effective communication and collaboration between teams, clear decision-making processes, and foundational capabilities already in place from previous investments. The company brought together dozens of mission teams, prioritized resources, and ensured transparency and accountability through regular check-ins and forums for design and architecture review. They also emphasized the importance of fast decision-making and clear escalation paths. Additionally, Intuit's previous investments in platform, AI, and data enabled the project's success. However, they acknowledged that they couldn't solve everyone's problem and made the system extensible for customers through customization options. Overall, the success of Genoese was a result of the company's processes, organizational culture, and foundational capabilities.
Centralizing AI capabilities: To effectively scale AI implementation in a company, centralize and extend existing capabilities, ensuring responsible data governance, compliance, and efficient development processes.
To effectively implement Generative AI (Gen AI) and Large Language Models (LLMs) at scale in a company like Intuit, it's essential to bring together existing capabilities, extend them as needed, and centralize their management under one umbrella. This includes well-defined data governance strategies, responsible AI training, and compliance with various regulations. Intuit already had a solid foundation with a unified data architecture, but the nature of the data changed from structured to mostly unstructured. To accommodate this shift, new controls and structures had to be developed for unstructured data. The existing authorization systems were leveraged, but fine-grained access controls needed additional development. The key to acceleration lies in automating these processes and making them easier for teams to use, allowing them to focus on delivering business value to customers without worrying about governance and compliance. Intuit's success in implementing Gen AI and LLMs came from recognizing the importance of centralizing and extending their existing capabilities. By doing so, they ensured that data is handled responsibly, security is maintained, and compliance is met, paving the way for a more efficient and effective development process.
Data retention policies in Gen AI: Adding YAML files to an ongoing framework of security policies enables quick implementation of data retention policies in Gen AI systems, allowing for consumer-friendly experiences and faster experimentation.
Implementing data retention policies in Gen AI systems can be done quickly and easily, allowing for a more consumer-friendly experience. The process involves adding YAML files to an ongoing framework of security policies, making it a low-code or no-code solution. However, despite the fast pace of advancements in Gen AI, companies are recognizing the need for even faster experimentation and are opening up more avenues for teams to explore new technologies, as long as sensitive data is not compromised. This approach allows the core team to focus on prioritizing and embedding the most valuable learnings into the operating system. Rapid experimentation and learning from customers are currently the top priorities for Gen AI developers, as they work to enable more capabilities for users. Additionally, the community on Stack Overflow was recognized for sharing knowledge, with a particular question on extracting metadata from MP3 files helping over 18,000 people.