Nearly every company in the Sequoia network is building language models into their products

The advent of large language models like ChatGPT has spearheaded a new era of innovation in Artificial Intelligence (AI). Companies are leveraging these Intelligent Agents to create Non-Human Workers or Digital Employees that can interact naturally with clients, streamline workflows, and even form relationships. A report by Sequoia Capital titled “The New Language Model Stack” sheds light on this groundbreaking trend, highlighting how various companies are integrating AI to elevate their offerings.

• A staggering majority of companies within the Sequoia network are integrating language models into their products, thereby transforming various sectors. For instance:

- Customer support and employee support chatbots are becoming more sophisticated.

- AI is redefining workflows in visual art, marketing, sales, legal, accounting, data engineering, search, grocery shopping, consumer payments, and travel planning.

• The technology stack central to these applications hinges on language model APIs. Key findings of the report include:

- 65% of the companies have applications in production, a significant increase from 50% two months prior.

- 94% are employing foundation model APIs, with OpenAI’s GPT models being the most popular (91%).

- 88% of the respondents believe that retrieval mechanisms, such as vector databases, are crucial for improving the quality of AI-generated content.

- 15% of companies built custom language models, signaling a growing interest in tailoring models to specific applications.

• Customization is a vital aspect as companies seek to apply language models in unique contexts. Three main methods of customization have been identified:

- Training a custom model from scratch.

- Fine-tuning a pre-existing model with domain-specific data.

- Using a pre-trained model and retrieving relevant context.

• There is an ongoing convergence of the stack for custom model training and that for leveraging language model APIs, with companies now exploring combinations of these technologies.
• The technology stack is evolving to be more developer-friendly, putting powerful AI tools into the hands of average developers, not just specialized machine learning teams.
• Trustworthiness is a critical aspect for wider adoption, with companies seeking better tools for data privacy, security, and monitoring model outputs to avoid errors or inappropriate content.
• Multi-modal applications combining text, voice, and images are on the rise, providing richer and more engaging user experiences.

This revolution signifies a pivotal moment in the evolution of AI technology, with Intelligent Agents and Non-Human Workers not just supporting but reshaping the business landscape. However, the technology is still nascent, and as it matures, there are immense opportunities for innovation and growth.

Reference: [1].