In a recent presentation at MachineCon 2023, Peeyush Dubey from TheMathCompany (MathCo) discussed the untapped potential of Generative AI (GenAI) and how it can be harnessed beyond the hype. He provided insights into the value chain, use cases, potential pitfalls, and accelerators of GenAI, offering a comprehensive overview of the current state of this technology.
The AI Opportunity
Dubey began by discussing the AI opportunity, emphasizing how AI is gradually outperforming human capabilities in various fields. He highlighted the potential of AI in enterprise settings, where it can be used to enhance operations across all sectors. The goal, according to Dubey, is to reach a point where AI performs better than humans in every organization, industry, and enterprise.
Dubey used Gartner’s definition to explain GenAI, describing it as a set of technologies that generate data, images, videos, or any other kind of data. He further explained the concept of large language models (LLMs), which use vast amounts of textual data to generate original artifacts. A subset of these LLMs, such as ChatGPT, is used for conversational purposes.
The Value Chain
The value chain of GenAI, as explained by Dubey, starts with hardware, typically provided by Nvidia, due to the high computing power required by these applications. This is followed by cloud platforms, closed-source models like OpenAI, open-source models like Dolly, and application enablers that form an abstraction layer on top of these models. Enterprises can then create their workflows using these enablers.
Use Cases and Implementation
Dubey suggested that enterprises should start with pre-built and hosted model APIs, such as OpenAI, for their initial use cases. As they scale and their use cases grow, they can move to open-source models or even build their LLMs from scratch. He emphasized the importance of selecting use cases where 100% accuracy is not needed due to the potential for hallucination and bias in AI models.
Dubey warned about potential pitfalls, including hallucination, bias, data privacy issues, and cost. He advised enterprises to be aware that their data will be stored in the closed-source LLM for some time and that the costs of using closed-source models can add up quickly.
MathCo offers accelerators that can help enterprises create templates for consulting, strategy, foundation, and tooling. Dubey suggested that the real potential of LLMs will be realized when they are integrated into the entire structure of an enterprise, from the foundation of engineering to the intelligence layer and the business value layer.
In conclusion, Dubey’s presentation provided a comprehensive overview of GenAI, its potential, and how it can be harnessed in an enterprise setting. He emphasized the importance of understanding the technology, selecting appropriate use cases, and being aware of potential pitfalls. His insights offer valuable guidance for any enterprise looking to explore the untapped potential of GenAI.