AI as the Catalyst for the Fourth Industrial Revolution – Insights from Professor Tien-Fu Chen, Vice Dean of the College of Computer Science

Leveraging Open-Source Large Language Models to Accelerate Project Timelines

In the early stages, companies typically focused their AI projects on applications such as image recognition and defect detection, aiming to reduce errors, improve product quality, and lower operational costs. However, today's GenAI projects are becoming key drivers for enhancing efficiency and powering future business growth.

Large enterprises see this technology as critical in boosting their market competitiveness. However, many need to pay more attention to the fact that training LLMs is highly complex, time-consuming, and costly. Most large companies begin by training their AI models, but completing the entire process-including inference, testing, and implementation-can take 1-2 years before yielding results.

The demand for computing power during the AI model training process is immense, requiring significant financial investment that only resource-rich companies can afford. "When developing GenAI models, companies should invest in training and inference separately, rather than attempting to do everything in-house, as it is extremely time-consuming and often fails to deliver the expected results," explains Professor Chen. "With the explosion of GenAI, many pre-trained AI models are now available on the market. Companies can leverage these models, combine them with internal data, and perform inferences on smaller-scale AI platforms, shortening AI project timelines. Most medium to large enterprises also have sufficient budgets to support this approach."

Currently, widely used open-source large language models include BLOOM, LLaMA3, and Mistral, which have already been broadly adopted. These models come in various parameter sizes, allowing companies to choose models based on their specific use cases. It is only sometimes necessary to use models with 175B parameters or more. Instead, companies can select models with appropriate parameters for fine-tuning or combine multiple smaller models using a mixture-of-experts (MOE) approach. This strategy reduces infrastructure costs and shortens AI model inference times.

Leveraging External AI Experts to Minimize Learning Errors and Accelerate Progress

Although open-source LLMs are emerging rapidly, companies can still establish their internal platforms for large language models. Recognizing the high level of interest in AI projects, businesses often need help with challenges such as a shortage of AI talent and limited resources. To address these issues, public cloud providers offer a variety of large language models for users to choose from.

Numerous information service companies on the market also provide one-stop solutions, including platform setup and consulting services. These solutions benefit small and medium-sized enterprises with limited budgets and human resources, allowing them to implement AI projects affordably and leverage AI technology to enhance productivity and competitiveness.

Professor Tien-Fu Chen pointed out that compared to the early stages of AI technology development, the emergence of ChatGPT, with its remarkable ability to respond fluidly and provide vast knowledge across various fields, has rapidly increased interest in GenAI among businesses. Companies now see GenAI as a way to strengthen their operational capabilities. However, as previously mentioned, the success of AI projects depends on the effective integration of three key factors: data, AI algorithms, and computing power. Ensuring that GenAI delivers the expected benefits relies heavily on how well these elements are coordinated, which could pose a significant challenge for businesses.

Considering market trends and companies' future development, many information service providers now offer one-stop AI project solutions. Companies can take advantage of external resources by starting with a consultancy approach using professional AI teams to implement projects on a small scale or in non-core departments. The most significant advantage of collaborating with external AI experts is that it allows companies to avoid substantial upfront investments before fully understanding GenAI technology, enabling more accurate assessments of potential benefits.

Additionally, most external experts have extensive experience in successful AI implementations, allowing businesses to gain insights into the finer details of project execution and identify areas of improvement, which can be highly beneficial for future AI initiatives.

Training Employees as AI Experts: A Challenge Companies Must Address

According to a report published by IDC, global spending on AI reached $432.8 billion in 2022, covering software, hardware, and AI systems. By 2026, global AI spending is expected to grow to $300 billion, with a compound annual growth rate (CAGR) of 26.5%, demonstrating the rapidly increasing demand for AI applications globally.

In this environment, the world is experiencing a surge in demand for AI talent. To address the challenge of recruiting skilled external talent, many companies are partnering with external research institutions to train their employees to be AI experts.

One of the most notable examples comes from several large electronic product manufacturers that have successfully used AI technology to drive digital transformation. To meet the demands of this transformation, these companies began collaborating with the Taiwan AI Academy in 2018, offering AI training programs for their employees with a focus on research and supply chain optimization.

This substantial investment has brought significant operational benefits. For example, some companies have adopted the concept of "machine doctors," using AI for visual inspections and "listening" diagnostics to assess the health of equipment and its components.

In intelligent healthcare, the accumulation of big data is the foundation of everything. Only with a rich and high-quality database can AI be endowed with predictive and decision-making abilities, ultimately achieving the effectiveness of precision medicine. (Image generated by ChatGPT)

Professor Tien-Fu Chen believes that the talent cultivation strategies adopted by leading intelligent manufacturing companies can be valuable lessons for other businesses, as they are part of an essential process. When companies recruit AI talent externally, they often face challenges such as a lack of industry knowledge and familiarity with the company's unique characteristics. These employees may need time to adapt and learn, and there is always the risk of being poached by other companies. Therefore, businesses must invest in training their employees to become AI experts.

Of course, some companies are concerned about new employees gaining access to sensitive information. However, this can be addressed by implementing a practical data access policy that balances security with the company's future growth.

It is undeniable that implementing GenAI projects helps improve work efficiency. Chen suggests that business leaders should not expect drastic changes overnight but focus on achieving incremental improvements of 10% to 20%. By making gradual changes, companies can find the most suitable transformation strategy. NYCU's myLLM industry-academia alliance platform offers a great starting point for businesses. Companies can quickly establish their LLMs and then evaluate whether to further invest in developing their own GenAI solutions to achieve more excellent digital transformation results and benefits.

Please refer to the original text (Mandarin) for details.