The misconceptions around ChatGPT and the potential threat it poses to Google and other search engines

Since its public unveiling at the end of 2022, many have speculated that ChatGPT is the ultimate route for Microsoft to gain market share and overtake Google as the leading provider of search. In fact, some have even gone as far as saying that it will be a Google Killer, ending its supremacy of search engines online. However, the idea of generative AI making search irrelevant is a misunderstanding of what this technology genuinely represents. 

If we look at how Google has launched Bard, its alternative to ChatGPT, it’s clear that generative AI is not a threat to search but rather an enhancement. Marketed as a complement to search, Bard represents Google’s entry into the generative AI market and its chance to rewrite the narrative around this technology. With ChatGPT and Bard taking the internet by storm, this distinction is crucial for organizations. While generative AI is powerful, complementing it with search greatly enhances its power and versatility, and may be the perfect solution that businesses have been searching for to gain a competitive edge.

Why are businesses holding back?

Initial trials of ChatGPT have allowed users to quickly complete and automate mundane tasks. The potential of this has swept through the corporate world. However, many business leaders are uncertain about how to implement generative AI models.

Most users of the technology acknowledge that whilst ChatGPT is great at natural language, creativity, conversing, and summarizing, it struggles in fact-based and nuanced contexts. This is because ChatGPT is a generative AI, trained to generate text based on language patterns and create compelling prose and confident, convincing arguments. The content is based on the probabilities of words in language rather than on an understanding what the words actually represent. It has no understanding of the real world. Consequently, the results struggle in complex environments and may hallucinate; that is, they can’t be relied on to convey accurate information at all times. When combined with the knowledge that these models have no knowledge of the last year or two, and that it’s difficult to validate the models’ sources, it’s easy to see why enterprises are hesitant to adopt them for many fact-based use cases. 

But this is where search makes all the difference.

The impact of search

Where generative LLMs (GLLMs) create a new response, search is about retrieval of information that already exists. You request pre-existing information from an insight engine, and it surfaces the most relevant content. GLLMs can appear very similar, as they present relevant material that reflects their training. In actuality, predicting which words to display about a subject is very different from showing content written on that subject.

However, starting with search and feeding the results of a search into a GLLM eliminates these shortcomings. Using search as the information source, rather than the LLM serving as its own source, ensures that responses are constructed from accurate, up-to-date, and traceable information. Working in this way puts each tool’s strength to good use: the knowledge comes from the most relevant information found by search, while the phrasing comes from the GLLM. 

The final result is an expression of accurate, traceable, up-to-date information presented in the way we most like to communicate: natural language. 

Since the accuracy of the response is directly related to the quality of information fed into the GLLM, search is now more important than ever.

But that’s not the best part

Not only does search provide better quality information, but it is also the most effective way to make a GLLM aware of enterprise-specific information.  ChatGPT and other GLLMs (GPT-4, Bard, LLaMA, etc.) are trained on public content sourced from the internet, and they have no awareness of the knowledge contained within an organization. Meanwhile, training or fine-tuning these models on enterprise content is difficult, can be incredibly time-consuming, and expensive. 

The ideal solution is to utilize enterprise search, which knows everything about a company, to provide relevant organization-specific knowledge to a GLLM. As enterprise search has broad and secure access to all the corporate repositories, content, and institutional knowledge, the GLLM gets the most relevant source material for its answer, from all enterprise content, regardless of source, format, or language. It also ensures security as employees only see the information that they have permission to access. 

The result of this combined approach is an enterprise-ready GLLM that is:

• Aware -- as it includes your company’s knowledge, not just public content
• Accurate -- because the information comes directly from your corporate content, not the model, for fact-based summaries without hallucinations
• Transparent -- with explicit links to sources so knowledge is traceable
• Current -- the content generated uses the most up-to-date information

Looking Ahead

GLLMs, like ChatGPT and Bard, are changing how we interact with information, and they’re going to continue to make headlines. But, as impressive as they are, it’s important to remember that we’re just at the beginning of this AI revolution. Over the next few years, we’re going to continue to see the development of even more capable models that will enable new applications that we can’t even begin to predict yet.

But without a foundation of concise, accurate, and relevant knowledge, GLLMs will struggle to meet their potential, particularly for the enterprise. Intelligent search is an invaluable tool to bring GLLMs to the enterprise and make them usable in almost any language-oriented application. Combining the power of intelligent search with GLLMs like ChatGPT means employees can now converse with their content, increasing effectiveness, accelerating innovation, and even boosting employee engagement.

Photo Credit: Vlue/Shutterstock

Ulf Zetterberg is Co-CEO at enterprise search provider,Sinequa. Ulf has over 25 years of experience in leadership for global enterprise software companies. He co-founded Seal Software, the first to use an AI-powered platform to add intelligence, automation and visualization capabilities to enhance the management and utilization of contract data, in 2010. He is also an investor, advisor, and board member to several other businesses, with a primary focus on software and data analytics.