Speaking at a technology forum in Shenzhen today (15 November), Baidu CEO Robin Li stated that businesses could be wasting resources in develop large language models (LLMs) before any viable business cases are found.
Speaking on the behaviour of tech companies in China, Li stated that he had observed a trend for companies stockpiling AI related hardware and building huge data centres to house the training data needed for LLMs.
Chinese AI startup 01.AI has recently reported that it stockpiled over a year’s worth of Nvidia chips in anticipation of further US import sanctions on sensitive technology.
“A large language model itself is a basic foundation akin to an operating system, but ultimately developers need to rely on a limited number of large models to develop various native applications,” Li stated as reported by Reuters, “Therefore, constantly redeveloping foundational large models represents an enormous waste of social resources.”
Beatriz Valle, senior analyst at research company GlobalData, agreed with Li’s sentiments but stated that this waste of resources was not limited to China.
Whilst these problems may have been exacerbated by China’s specific demographic, economic and social conditions, Valle stated that countries around the world stand at a “critical juncture” of AI and LLM development. Mainstream uptake of LLMs by Western consumers, Valle points out, will lead to higher pressure on Western companies to create long-term viable solutions to monetise AI.
Whilst many companies race to create their own generative AI chatbots, Valle believes that in the long-term only a handful of LLMs will become the industry standard.
“Smaller custom models, with fewer parameters and trained on proprietary enterprise data, will be the norm, rather than black-box style large language models,” she concluded.
Nathan Selby, client portfolio director at marketing services company Active Profile, stated that businesses were in a race to at least be seen as AI-focused without long-term considerations. Too often, he stated, these plans are not considering the potential repercussions of their AI projects. The cost of maintaining an LLM is often ignored for the face value of being seen as forward-thinking.
“Our strategic advice [to businesses looking to harness AI] would be to think really long and hard at the ‘why’ in ‘why are we doing this’?” he said.
GenAI is currently seeing an uptake in almost every sector since the public release of OpenAI’s ChatGPT last November.
In a 2023 Q3 GlobalData survey measuring the rate of technological disruption, AI was by far considered the most disruptive technology by respondents across sectors. Around 76% of participants stated that AI posed either significant or slight disruption to their industry, and a further 15% anticipated that it would soon disrupt their industry in the next 12 months.
GlobalData graduate analyst, Emma Christy, says of this disruption: “By 2024 every company will need to be an AI company, particularly as AI becomes more powerful and accurate.”
In Christy’s opinion, businesses should focus on finding viable use cases for AI before investing, but this does not necessarily mean that turning to AI would be a waste of resources.
By as early as the next two years, she estimates that apparent use cases for AI will become progress and that AI will be a business necessity for companies wishing to have a competitive edge.
By 2030, GlobalData estimates the total global market of AI to be worth over $900bn.
