Exploring opportunities in the generative AI value chain: What to Watch in AI

Exploring opportunities in the generative AI value chain: What to Watch in AI

The field of generative AI is rapidly evolving, with applications built from fine-tuned models standing out as particularly promising. When it comes to generative AI applications, we can broadly categorize them into two groups. The first group consists of companies that use foundation models as is within their applications, with some customizations. These customizations may include creating a tailored user interface or adding guidance and a search index for documents to enhance the models' understanding of customer prompts and improve output quality.

However, it is the second category that presents the most attractive opportunities in the generative AI value chain. These applications leverage fine-tuned foundation models that have been fed additional relevant data or had their parameters adjusted to deliver outputs for specific use cases. Fine-tuning foundation models requires less data, costs less, and can be completed in days, making it accessible to many companies. By fine-tuning models, companies can create proprietary data through feedback loops driven by an end-user rating system, such as star ratings or thumbs-up, thumbs-down ratings.

As the demand for generative AI applications grows, dedicated generative AI services will emerge to help companies fill capability gaps. These services will assist companies in building out their experiences, navigating the business opportunities, and handling the technical complexities associated with generative AI.

While much of the generative AI trend has been focused on consumer applications, there is a growing emphasis on serving enterprises. Companies like Glean, Lamini, Dust, and Lance are directly targeting enterprises by building products that incorporate internal data and adhere to corporate guidelines. This shift towards enterprise-focused generative AI applications is driven by the need for more accurate representations of the world and the increasing sophistication of attacks.

In today's digital landscape, the number of cyber attacks is skyrocketing, with attackers becoming more sophisticated. Just as any student can use generative AI models to write an essay, these models can also be used to produce fraudulent messages that are grammatically perfect and personalized, posing a significant threat. Dust has developed a platform that indexes and embeds companies' internal data from various sources like Notion, Slack, Drive, and GitHub, enabling the development of generative AI-powered products. By leveraging fine-tuned models and proprietary data across multiple modalities, enterprises can create AI solutions that offer differentiated services, insights, and increased operational efficiencies.

Labelbox is another company addressing a critical challenge in AI development: simplifying the process of feeding datasets into AI models. Enterprises often struggle with incorporating their proprietary data into AI models, and Labelbox aims to streamline this process. By facilitating the integration of proprietary data, companies can leverage their unique datasets to enhance the performance and accuracy of their AI applications.

While many companies have incorporated AI as chatbots to enrich existing applications, there is a need for AI applications that go beyond augmentation and fundamentally change how we interact with products. AI has the potential to dramatically improve the user experience by transforming creative tools and reinventing product experiences. Lamini, for example, is an LLM (Large Language Model) engine that empowers developers to rapidly train, fine-tune, deploy, and improve their LLMs with human feedback. This approach allows for continuous improvement and refinement of generative AI models, leading to more impactful and innovative applications.

However, one of the key obstacles preventing enterprises from shipping AI applications to production is the lack of appropriate governance controls. Ensuring that applications understand what end-users are allowed to see, determining where inference is performed, and identifying the source data behind a model's output are crucial governance considerations. Glean has emerged as an enterprise-grade AI data platform and vector store that addresses these governance challenges. By integrating with an enterprise's internal environment and real-time data permissions, Glean enables enterprises to enforce governance controls at scale and confidently leverage their internal data for model training and inference.

In conclusion, the generative AI value chain offers exciting opportunities for companies to develop innovative applications. By leveraging fine-tuned models, incorporating proprietary data, and focusing on enterprise needs, companies can create differentiated services, enhance operational efficiencies, and improve the user experience. To succeed in the generative AI space, companies should consider the following actionable advice:

1. Invest in fine-tuning foundation models: Fine-tuning foundation models can deliver highly customized outputs for specific use cases while requiring less data and lower costs compared to training foundation models from scratch.

2. Embrace enterprise-focused generative AI: Enterprises present a significant market for generative AI applications. By building products that adhere to corporate guidelines and incorporate internal data, companies can cater to the specific needs of businesses and offer valuable solutions.

3. Prioritize governance and data control: As AI becomes more widespread, ensuring appropriate governance controls and data ownership is essential. Companies should consider solutions like Glean that enable them to enforce governance controls, understand data ownership, and confidently leverage internal data.

By considering these actionable advice and staying ahead of the trends in generative AI, companies can unlock new opportunities, drive innovation, and gain a competitive edge in this rapidly evolving field.