Published : 2 years ago, on 02/06/2023

Large Language Models are the basis of the new generation of AI engines, but private foundation models are the future

How foundation models can unlock unstructured data for enterprise

By Marshall Choy, SVP, SambaNova Systems

A Fourth Industrial Revolution

The start of 2023 has seen generative AI explode into the mainstream, with ChatGPT introducing the transformative power of this technology to a broad consumer audience. Big tech firms are taking steps to turn theory into practice, lifting AI projects out of the sandbox and into the real world. Since ChatGPT was released, a wider audience has been exposed to the emergent capabilities of Large Language Models (LLMs) to achieve results beyond the scope of their initial creation. The floodgates are creaking open, and the fourth industrial revolution is beginning.

Revolutions don’t, however, happen overnight. Behind this upsurge in interest, AI innovators have been methodically fine-tuning AI models for their breakthrough moment. Now that that moment has arrived, the appetite for AI – from both consumer and enterprise – has skyrocketed. Industrial revolutions, however, don’t take place in the consumer realm. Industry is the engine that fuels societal change, and it is through industry applications that AI will bring the most significant societal benefit.

ChatGPT is just the tip of the iceberg for generative AI capabilities, and enterprise-focused AI vendors are leading the way in bringing additional capabilities to market. Enterprise-ready AI is not the next step following ChatGPT; it’s the very foundation on which future success will be built and has proven its reliability and readiness to supercharge businesses.

Enterprises leading the way 

Enterprise continues to lead the way in implementing AI to significantly improve customer services and operational efficiency. For example, the banking industry is already adopting advanced language models to consolidate AI sprawl into high-performance foundation models. In addition, banks have recognised that they’ve been sitting on troves of unstructured data such as emails, chat logs, voice call recordings and financial reports. As a result, we’re increasingly seeing more and more banks put AI foundation models in place to gain trapped insights from this data.

Many businesses have employed conversational AI in forms like conversational AI chatbots for many years already. However, in 2023, we have seen AI leap into the domain of genuine ROI for businesses, and the potential use cases for enterprise expand beyond what was previously thought possible.

To get the most out of AI deployment, businesses need a considered long-term strategy and the expertise and knowledge base to fine-tune their AI model. Traditionally, new technologies are judged on their ability to deliver short-term ROI. However, with the current gold rush on generative AI, this can lead to the confusion of multiple models being adopted for different use cases simultaneously, limiting scalability and, ultimately, restricting their effectiveness.

Foundation models are the basis of generative AI and enterprise success

LLMs form the basis of a new generation of revolutionary AI engines. With ChatGPT, they’re showing their ability to comprehend and generate coherent language. LLMs are already being implemented across industries, generating insights from vast swathes of unstructured data. However, foundation models, the basis of generative AI and LLMs, are where the real value for enterprise can be found.

LLMs, like the GPT-based model that powers ChatGPT learn from enormous quantities of data. Unfortunately, this can lead to inaccuracies in disseminating flawed data, requiring more privacy and security for enterprise applications. Recently, we’ve seen organisations like Goldman Sachs instruct their employees to stop using the public ChatGPT model due to security concerns, and inaccuracy is simply not an option for enterprises looking to generative AI to optimise their operations.

Tailored, enterprise-ready generative AI models can avoid these problems of inaccuracy when finetuned with the firm’s own data. Businesses can reduce their tech sprawl by centralising functionality into a solitary model by utilising foundation models. Private foundation models with a business’s own data can enable companies to retain control over how that data is stored and used. Control over data is retained, and the best infrastructure for achieving strategic objectives can be utilised, both of which are top concerns for enterprises.

It’s an inspiring time to be at the forefront of generative AI as the AI gold rush kicks off in earnest. However, bringing these groundbreaking capabilities to enterprise and turbocharging business efficiency is the most exciting.