This article first appeared in Digital Edge, The Edge Malaysia Weekly on February 27, 2023 - March 5, 2023

Searching for information has become a more visual experience, thanks to generative artificial intelligence (AI), which represents one of the most significant evolutions on the internet. As the search process transitions from merely presenting users with relevant links to delivering relevant answers directly in the results page of the search engine, the user experience is enhanced.

The rollout of ChatGPT by AI research and deployment company OpenAI in November last year caused a frenzy over generative AI, with investments in the sector skyrocketing 425% to US$2.1 billion last year from 2020, according to financial data and software company PitchBook.

While the rollout of generative AI offerings could lead to shifts in search market share, it is still early days for any kind of transition to take place, says Citi Global Perspectives & Solutions (Citi GPS) in its latest report. According to its “Generative AI, ChatGPT and search” report, consumers tend to be “sticky” when it comes to their technology use, meaning it will take time for them to change their habits.

“It may be some time before this shift is evident. Every one percentage point change in search [market] share equates to around US$2 billion of search revenue,” the report adds.

Generative AI is an existing tool already used in applications like email via smart compose. However, its use has been increasingly ramped up due to the growth in model size and complexity. For instance, Large Language Models (LLMs) can analyse huge amounts of data sets. 

LLMs predict what word or series of words comes next in a sentence. It is also used to improve speech recognition software by reducing the likelihood of illogical words being transcribed by the software. This has led to them being used by online support chatbots.

The development of OpenAI’s three generative AI models — ChatGPT, Codex and DALL-E — has become an inflection point in the industry due to the improvements in computing, data availability and the public’s ability to test and refine the models. ChatGPT, in particular, is a user-friendly application of LLMs, proving that progress in the technology is indeed being ramped up. 

OpenAI’s GPT-3, an autoregressive language model, is one of the greatest leaps in the abilities of LLMs in recent years. GPT-3 performs better at tasks it is trained in. Through simple scaling of the learning data set and computational power of the model, it was able to perform better at tasks it was not explicitly trained in.

As such, generative AI could lead to shifts in search market share, with major players in the space already working on generative AI and LLM offerings. This could lead to a significant change in the landscape over the near to medium term, states the Citi GPS report. 

Earlier this month, Microsoft announced its Bing search engine and Edge browser would be running on a new Open AI LLM that is customised specifically for search. It takes key learnings from ChatGPT and GPT-3.5 This integration would allow for the creation of content based on personalised results, complete answers to questions summarised with links across the web, chat for follow up and specific queries, and expanded search with a 1,000-character text box. 

Bard, Google’s first conversation-based AI model, was unveiled this month as well. Users can ask open-ended questions and explore answers related to the topics. They can then follow up with related requests to engage the model further. Alternatively, they can begin a new query. 

Meanwhile, internet companies in China are developing products to incorporate ChatGPT-like features into their services. 

LLMs are currently restricted by the available computing power. As the size of LLMs grows, there has been an increase in the need for computing power to train them. Generative AI is also expensive in terms of development and service.

Furthermore, the reliability of ChatGPT’s results is an open question. There could be the potential risk of misleading answers. As ChatGPT is a pre-trained model, its content is highly dependent on input and refinement of the model. Using traditional search, users are directed to the original source with limited liability on the accuracy of the contents. 

Despite the challenges faced, modern LLMs are held in high regard for their excellent ability to understand and generate text as well as internalise massive amounts of real-world knowledge during initial training. 

Save by subscribing to us for your print and/or digital copy.

P/S: The Edge is also available on Apple's AppStore and Androids' Google Play.