Sequoia Research Reveals Emerging Trends in Integration of LLMs

The Sequoia study reveals a remarkable surge in the integration of LLMs into various products throughout its network. The application of LLMs has expanded from autocomplete features for coding to enhanced chatbots for customer service, infiltrating nearly every aspect of business operations. The permeation of AI is also leading to radical transformations of entire workflows in sectors as diverse as visual art, marketing, sales, contact centers, and beyond. A rapidly growing trend in the deployment of LLM applications is the adoption of language model APIs, retrieval mechanisms, and orchestration tools. However, open-source usage is also on the rise. Some key findings from Sequoia's analysis include:

• 65% of the surveyed companies have applications in production.
• 94% are using a foundation model API, with OpenAI's GPT proving the most popular.
• 88% upheld the importance of retrieval mechanisms in improving results and reducing inaccuracies.
• 15% are crafting custom language models from scratch or leveraging open-source alternatives.

Sequoia's report underlines the fact that while general language models are powerful, they often lack the differentiation or specificity required for certain use cases. Companies are demonstrating a keen interest in customizing language models to meet their individual needs. This includes leveraging a wide variety of data, from developer docs and product inventory to HR rules and user-specific data. Model customization can currently be accomplished in three ways: training a custom model from scratch, fine-tuning a base model, or using a pre-trained model with the retrieval of pertinent context. As the methodologies for customization evolve, a merger of LLM API and custom model stacks is expected. The introduction of language model APIs has democratized access to robust models, sparking the development of more developer-centric tools. More and more developers are turning to LangChain to build LLM applications, thanks to its ability to simplify the process by addressing commonly encountered issues. Data privacy, output quality, and security are crucial aspects that need to be addressed for LLMs to gain widespread acceptance. Numerous companies, particularly in regulated industries, are actively seeking software solutions that can enhance data privacy, segregation, security, copyright, and monitor model outputs. As these demands are met, the adoption rate of LLMs is set to soar. The future promises a wealth of potential for multi-modal language model applications, with companies already starting to combine different generative models. The fusion of text and speech generation can pave the way for a new generation of chatbots, delivering a more fluid conversational experience.