CONNECT WITH US

Former Google director sees long-term needs for Taiwan to cultivate large language models

Ines Lin, Taipei; Misha Lu, DIGITIMES Asia 0

Credit: DIGITIMES

As ChatGPT gains traction, Taiwan's industry, government and academia have declared the intention to develop large language models (LLM) one after another. In this regard, Lee-feng Chien, former managing director of Google Taiwan, believes that it is difficult for Taiwan to compete with major international players in the field of artificial intelligence (AI) in the short term. However, there is still a need to develop related technologies and talents, especially natural language processing-related talents which are relatively scarce.

On March 22, Taiwan Internet and E-Commerce Association (TiEA) held a member meeting and industry forum. Chien was the honorary vice chairman of TiEA. Many new start-up companies, stock exchange innovation boards, and venture capital representatives also attended the forum to share observations on industry trends.

Regarding Nvidia and cloud service providers jointly launching AI as a service, Chien pointed out that NVIDIA intends to be the Microsoft of the AI industry. When its hardware development reaches a certain level, a software development platform would be needed to support hardware. When it comes to developing software development platforms, the risks are lower but the margins are higher. For GPUs to be successful, Chien said that a complete toolchain will be necessary as well, as it takes a long time for developers to learn new technologies.

On the rumored new wave of layoffs from Amazon and its impacts to cloud business, right at a point when AI's popularity reaches a new height, Chien observed that the economic shock since 2022 has resulted in reduced expenditures, and most cloud service contracts are signed on an annual basis, usually finalized at the end of the year. This explains why Microsoft, despite its good position, still announced 11,000 layoffs at the beginning of the year.

Chien added that by the end of 2022, large enterprises around the world should have made considerable adjustments to their budgets. This phenomenon has the greatest impact on large B2B companies such as AWS and Microsoft.

Taiwan's National Science and Technology Council has announced that it will lay the groundwork of a generative dialogue engine by the end of 2023 through public-private cooperation. MediaTek, Academia Sinica and Academy of Chinese Education also worked together and have recently released a traditional Chinese language model, open for testing on open source websites. Commenting on the endeavors, Chien sees more long-term benefits compared to short-term ones.

After all, many language models developed already cover Chinese, and different language systems can be regarded as a token system, posing no significant challenges to the development and implementation of AI models. In the long run though, it is still necessary for Taiwan to cultivate large language models and related talents, as it is unknown if special needs will emerge in the future that the current leading service providers cannot address.