While the inaugural AWS Large Language Model (LLM) League held in Singapore last year showed how anyone can unlock the full potential of artificial intelligence (AI), this year’s expanded edition, the Regional LLM League, brought the stage to young people across Southeast Asia to democratize access to AI through fine-tuning large language models (LLMs), demonstrating the synergy of those skills when talent converges.
Co-organized by Amazon Web Services (AWS) and AI Singapore (AISG), the first regional expansion of the Regional LLM League attracted 1,300 students from highereducation institutions in six countries: Vietnam, Indonesia, Malaysia, Philippines, Singapore and Thailand.
The competition kicked off in January and culminated in the Grand Finals in May, as part of the AI Student Development Conference (AISDC). Hosted by AI Singapore (AISG), the AISDC brings together industry leaders and emerging talent in the AI space to drive innovation and shape the future of technology. The event showcased real-world AI applications, career opportunities and thought leadership through a series of panel discussions, workshops and exhibitions, with the aim of inspiring and equipping the next generation of AI talent in Singapore and the region.
Mr Koo Sengmeng, Director of AI Talent Development, AI Singapore, said: “LLM League originated from a bold idea by the fearless innovators at AWS – to gamify the refinement of LLM models and make them more engaging and accessible to everyone. AI Singapore is proud to partner with AWS to bring these essential 21st century skills to life in a fun and accessible way, not just in Singapore but across Southeast Asia.”
“The LLM League is a clear demonstration of our vision to make digital skills accessible to learners from all backgrounds, especially students. I am delighted to see so many young talents coming together to master cutting-edge AI technologies,” said Elsie Tan, Country Manager, Global Public Sector, AWS Singapore.
To date, AWS has trained more than 31 million students in more than 200 countries and territories in cloud skills through free training programs, thereby opening the door of opportunity for underserved communities, actively contributing to shaping the future of technology, said Ms. Elsie Tan.
A stage for students to showcase their creativity
Participating in the tournament, contestants are given Amazon SageMaker JumpStart credits to refine their Llama 3B models, with the ultimate challenge goal: surpassing the much larger Llama 70B reference model, in the form of a multiple-choice assessment.
Through an in-depth workshop on Generative AI (Gen AI) led by instructors from Gen-C (AWS Gen AI Learning Community), students are exposed to key concepts in LLM model development, including dataset management, prompting techniques, and evaluation methods, as well as agentic AI and responsible AI.
In just three weeks, they created more than 5,000 models—a clear demonstration of the program’s effectiveness, and a testament to the value of skill and fine-tuning strategy in developing an AI model, rather than just relying on its size.
Following a series of regional rounds and internal evaluations, the top students from six countries were invited to compete in the Regional LLM League finals, which took place at the Equarius Hotel in Singapore on May 29. Each student was evaluated on two main criteria: the accuracy of their model through multiple-choice questions, and their ability to fine-tune their prompts in real time to optimize the model’s response. The judging panel included experts from AISG and AWS, along with votes from the audience – and, crucially, from the AI itself. The evaluation was supported by a standardized scoring system developed by the Anthropic Claude 3.5 Sonnet model, to ensure fairness, comprehensiveness, and objectivity throughout the competition.
Blix D. Foryasen, a computer science student from the Philippines, won first place. Using model distillation — a technique for training a small model using the output of a larger “teacher” model — Blix experimented with different teacher models to build a highly accurate training dataset. As a result, his model outperformed a larger model in the multiple-choice test.
“Evaluating the model’s response to data management and inference generation was both an exciting and memorable experience for me,” Foryasen said. “This challenge clearly revealed both the capabilities and limitations of AI, and also affirmed the irreplaceable role of humans in controlling bias and limiting misleading outputs, especially in the context of increasingly popular AI applications.”
Coming in second was Kim Seokyung from Vietnam, a software engineering student experiencing LLM fine-tuning for the first time, proving that the learning journey is not far away.
“I started from scratch, but within a few weeks I was able to build a refined model ready for real-world use,” she says. “It was a stressful but incredibly rewarding experience, and I am now confident in applying this skill to real-world use cases that are in increasing demand across a variety of industries.”
Building personal skills for the benefit of the community
One of the key goals of the Regional LLM League is to provide hands-on learning experiences in real-world environments – the model development scenarios that contestants must solve all simulate situations in professional AI environments.
Students are also provided with the Amazon SageMaker JumpStart toolkit to train models, test inference, and simulate deployments, allowing them to not only develop and hone their technical skills, but also gain exposure to real-world application environments.
The success of this first regional tournament has demonstrated the scalability of a community-based technology upskilling model. As an initiative of AI Spring Communities under the AWS AI Spring Singapore program, the tournament serves as a model for the widespread adoption of AI education at scale, especially in regions with disparities in education systems and access, as well as in communities that are underserved or underrepresented in various sectors.
By integrating Gen AI training with challenges and community impact goals, initiatives like these are gradually bridging the gap between academic passion and a career in AI. Participating students not only gain a portfolio of real-world projects and experience working on small- to large-scale models, but also gain the confidence to enter internships, pursue research, or develop their own AI projects.
As Gen AI becomes a strategic pillar of digital transformation, the need for a diverse and inclusive talent pool is more important than ever. The Regional LLM League is a clear demonstration that, with the right tools, mentorship, and opportunities, anyone can make a positive contribution to the future of AI development.
Source: https://nhandan.vn/giai-dau-aws-regional-llm-league-dao-tao-tai-nang-tri-tue-nhan-tao-the-he-moi-post893496.html
Comment (0)