Dr. Zhang Hongjiang, advisor to WIZ.AI and a famous computer scientist, shares his perspectives on the recent trends in AI and Large Language Models (LLMs), the transformative impact of foundation models like ChatGPT and GPT-4, and their implications for AI and generative AI adoption primarily in Southeast Asia in this interview with WIZ.AI. The discussion also explores the challenges, risks, and opportunities confronted by AI companies and the crucial role of talent development and AI governance. Read on to find out more.
- The adoption of ChatGPT by millions of users will be a key milestone in AI history. In ten to fifty years time, this year will be seen as the year that saw and realized the full potential of AI.
- Foundation models will not just be built in the US , but also from other regions like Southeast Asia. In Southeast Asia, Singapore would become a regional hub for new AI technology due to the density of AI talent and its adoption in Singapore.
- The current AI wave is different from previous ones as foundation models will require a much bigger resource pool than before for training.
- AI-based companies should invest their resources in addressing AI parsing and ensuring safety from potential risks. System failure and misuse are two such risks that companies have to tackle for enhancing security.
- This AI wave would replace jobs as well as create new opportunities. Therefore, retraining people will become critical. One of the programs from AI Singapore is focused on training non-technical staff in local traditional companies to equip them with AI and prompt engineering skills.
- Currently robots can only perform a specific task. But, in the future, robots equipped with LLMs and multimodality models can act autonomously based on observations.
- In the future, not just big enterprises, startups focused on technology could also become foundation model providers.
WIZ.AI: What do you think are some large language model trends and AI advancements happening around us?
Dr. Zhang Hongjang: In the last twelve months, the entire world has experienced a rapid change in AI technology. We witnessed the power of foundation models, ChatGPT, and GPT 4 in particular, changing how people look at AI, the future of AI, and how AI changes our societies, the way we do our businesses, and how we interact. If you look at how fast ChatGPT is being adopted by over 100 million users, it’s really a milestone in AI history. And I think in ten, twenty, or fifty years’ time, we can say this is really the year that we saw and realized the full potential of AI.
WIZ.AI: Advancements in AI are happening not just in the US. They are taking place in Europe, Asia, Southeast Asia, and other parts of the world. What are your views on the differences in AI trends in these regions, especially in Southeast Asia?
Dr. Zhang Hongjang: Building foundation models is not limited to the US. It is happening in Europe, the UK, China, and definitely, also in Southeast Asia. In Southeast Asia, for instance, Singapore has assembled a good set of AI talent at the National University of Singapore and Nanyang Technological University. I am impressed by the density of AI talent in Singapore and how quickly people learn and adopt this new technology, from government to private enterprises and from big multinational corporations to small startups. I am surprised by how fast and effective the Singapore government promotes AI technology.
While looking into the potential impact on society, I see Singapore as a regional hub for new AI technology and new scenarios to apply those technologies. Looking forward, Singapore will continue to play a critical role in AI adoption in Southeast Asia. Southeast Asia is a fast-growing economic region. There are several opportunities to apply foundation models and fine-tune them to particular scenarios that fit various countries, industries, and commercial sectors. In doing so, Singapore-based AI companies like WIZ.AI will have a lot of opportunities to enhance the importance of artificial intelligence.
WIZ.AI: How is the current AI wave different from the previous advancements in AI technology, and what are the changes brought forward by foundation models during this time?
Dr. Zhang Hongjang:This AI wave is a little bit different from previous ones. The reason is that the foundation AI models require a much bigger resource pool than before. They will also need CPU resource cards and pre-trained models for performing alignment training. This requirement for resources will continue in the future as well for training AI models. However, a few things are the same for Southeast Asia and other parts of the world. They include how we find those resources, aid companies access them, and help startups. These are some things we need to put a lot of effort into.
The foundation model has brought a change that we didn’t see in the past. In the past, if you wanted to develop a service or application scenario, let’s consider it as a medical application to look at X-ray pictures, you will need to do a vertical model for that particular scenario. But today, with large language models, especially with multimodality models, you can use a foundation model to fine-tune them to align to a particular scenario so that there is no longer the need for a big AI team in a company. And using your data to fine-tune and align is a paradigm shift in the AI application department. On the one hand, we know in most applications, you don’t need to develop new models. This could accelerate the AI application development and adoption of AI models to help improve productivity and foster the creation of new AI-powered apps for consumer services.
WIZ.AI: How do you think AI-based companies like WIZ.AI can strategically address the linguistic and cultural diversity in the SEA region by fine-tuning and aligning foundation models?
Dr. Zhang Hongjang: Given the demographic diversity in Southeast Asia, with multiple languages, cultures, and traditions, it is critical to align and fine-tune models to fit local cultures. This provides companies like WIZ.AI opportunities to play a role in adopting AI for customer services, corporate services, and more. However, it goes beyond customer service and could help improve the company’s productivity to consumers and to their customers. Thus, WIZ.AI, as an AI service provider, can provide tons of value to local companies. Even then, there are challenges to ensure we preserve the privacy and security of the data. Currently, there are new algorithms or systems that could protect the privacy and integrity of data and improve security.
WIZ.AI: What is your suggestion to generative AI solution providers like WIZ.AI for improving AI security?
Dr. Zhang Hongjang: AI technology and application companies should all spend a significant portion of their resources and efforts addressing AI parsing. This is like working with nuclear energy and nuclear power stations, where a large amount of R&D effort has to be on safety. In the past, AI’s power was far away from doing any harm. But, today, we have come to that point. You can categorize the risk into different types. System failure is one of them. This has been a persistent topic ever since the start of IT technology. What if the system fails? This would impact the government, students, and everyone in general. AI and LLMs are largely exposed to this risk.
The next category is misuse. When you don’t follow the regulatory rules and ignore the principles, that is misuse. There could even be rogue use of AI designed for evil purposes. This can also come under misuse. Therefore, we need governance and regulations for AI and generative AI. Those of us considering the technical aspects should be aware of the risks. But, working from a technological aspect, we should know how to discover and handle these risks.
WIZ.AI: Will this AI wave replace jobs? What are your thoughts?
Dr. Zhang Hongjang: Every new technology will bring a lot of changes to our society and industry. This means that some jobs will be replaced. Hopefully, it will also produce new job opportunities. Therefore, people can be retrained to take up these new job opportunities. But this is a much larger issue than technology alone. I recently conversed with individuals from the AI Singapore program.
One of their activities is training the non-AI technical staff from local traditional companies. These training programs equipped them with AI and prompt engineering skills and a good understanding of LLMs. These initiatives have helped those companies in adopting AI. These programs could benefit every Southeast Asian company, especially the traditional companies. The regional economy can benefit by training/retraining and providing employees from these companies with AI skills. AI is an advanced technology that can be hard to grasp. But, the foundation model has made it easy to use this technology, lowering its bar.
WIZ.AI: How far do you think we are from the era of AGI and what is your greatest hope for AGI?
Dr. Zhang Hongjiang: We have seen a spark of AGI from GPT4 and the newly released multimodality models. I think they will change the industry and how we develop software and future models. If you have a co-pilot, helping you with coding and design, it will accelerate your development process. Today, LLMs can handle most of the native language processing problems. I also see multimodality models, combined with vision and language, revolutionizing the software and robotic industry. So today, we notice many robots that can only perform one task. They are really hard-wired and precoded. But, in the future, once we equip those robots with LLMs and multimodality models, we can give them real brains. This will allow robots to act autonomously based on observations and your requirements and requests. This shift will transform robots into autonomous self-planning and self-action robots.
Of course, it will pose potential risks. We have just moved from a traditional information age to an intelligence age. In the future, we will move into an age of autonomous agents. We will have many software agents helping us in various ways, and they will work with us. They would work as co-pilots or autopilots. Again, this needs developments with a lot of importance on safety and risk management. Overall, I am very optimistic that not only can we develop the best technology in fast-moving areas, but we should also be able to find solutions to manage risks.
WIZ.AI: Do you think that in the future the big players will continue to remain as the providers of foundation models?
Dr. Zhang Hongjiang: No. It is not necessary. Open AI was a small startup and nonprofit two to three years ago. But, now, they are the most important players in the AI and generative AI space. So, I don’t think that it will be just the big players, who will have the opportunity to become a foundation model provider. It could also be those startups that are focused on technology. There are a few of them. Some of them will succeed for sure. Hopefully, in Southeast Asia, we can see one.
About Dr. Zhang Hongjiang
Zhang Hongjiang is a Chinese computer scientist and executive. He is an advisor to WIZ.AI. He has served as Kingsoft’s CEO, Microsoft’s managing director, Advanced Technology Center (ATC) and Chief Technology Officer (CTO) of Microsoft China Research and Development Group (CRD). He was elected to the National Academy of Engineering as part of his technical contributions and leadership in the area of multimedia computing in 2022. Learn more about Zhang Hongjiang from here.