Towards Building Multicultural and Multilingual Safe Large Language Models - 東京カレッジ

Towards Building Multicultural and Multilingual Safe Large Language Models

When:
2024.11.11 @ 10:00 – 11:00
2024-11-11T10:00:00+09:00
2024-11-11T11:00:00+09:00
Towards Building Multicultural and Multilingual Safe Large Language Models

Download the report on the webinar here.

Download the report on the Singapore AI Safety Red Teaming Challenge here.

Finished
Zoom Webinar
Date(s) Monday, 11 November 2024, 10:00-11:00 JST
Venue

Zoom Webinar (Register here)

Registration Pre-registration required (If you cannot attend on the day but would like to be informed of the youtube link at a later date, please register)
Language Japanese
Abstract

As generative AI becomes more widely used, it is crucial for AI models to accurately reflect cultural and linguistic risks in different regions. Identifying harmful content specific to each culture must be continuously updated. This requires collaboration between AI researchers, social scientists, policymakers, and practitioners to form a global community for ongoing discussions.

 

One approach, known as “red teaming,” involves intentionally trying to break AI models by inducing harmful content, but current red teaming efforts are mostly Western-focused.

 

To address this gap, Singapore’s Infocomm Media Development Authority launched a Singapore AI Safety Red Teaming Challenge in November 2024, gathering experts in culture and language to enhance AI safety in specific regions. Japan is also involved in this initiative. This event will report on the project’s November meeting and discuss frameworks to sustain such communities, welcoming those interested in AI safety and governance.

Program

10:00-10:15 Opening Remarks and Overview: Arisa EMA (Tokyo College, The University of Tokyo)

 

10:15-10:30 Efforts Toward LLM Safety: Satoshi SEKINE (NII-LLMC/RIKEN AIP)

 

10:30-10:45 Overview of the CTF Challenge and Future Plans: Teresa TSUKIJI (Japan Deep Learning Association)

 

10:45-11:00 Comments and Q&A

Contact tg-event@tc.u-tokyo.ac.jp
Organized by Tokyo College, The University of Tokyo; Institute for Future Initiatives, The University of Tokyo; Next Generation Artificial Intelligence Research Center, The University of Tokyo; B'AI Global Forum, The University of Tokyo; Japan Deep Learning Association
Supported by National Institute of Informatics, Research and Development Center for Large Language Models; Japan AI Safety Institute

Upcoming Events

The Meaning and Implications of Being a “Visitor-Centered” Museum (Lecture by Prof. Leslie BEDFORD)

イベント予定講演会/Lecture

Tuesday, 22 April, 14:30–16:00 JST

From "Being about Something to Being for Somebody" is the memorable title of Stephen Weil's contribution to the 1999 Daedalus volume on American Museums. Though decades old and not without challenges, the phrase continues to resonate within the museum field. In many ways it encapsulates the ways in which institutions have changed from privileging curator-defined information and institution-defined goals to promoting the visitor's experience and the broader community.
In her talk, Leslie Bedford, a longtime museum practitioner and professor of museum studies, will explore the meaning of Weil's phrase and how it has been implemented by museums, including those she has visited in Japan. Her talk will include a discussion of the online conversations she is facilitating with Japanese museum professionals and academics. And finally, she will raise the question of what "visitor centered" can mean today and in the future.

Designing and Scaling up Nature-based Markets (Lecture by Prof. Beatrice WEDER DI MAURO)

イベント予定講演会/Lecture

Wednesday, April 23, 2025 15:00-16:30 JST

Carbon and nature markets are struggling with low trust, high costs, and limited scale—falling far short of what’s needed. In this lecture, Professor Weder di Mauro presents a new market design co-developed with Estelle Cantillon and Eric F. Lambin. Jurisdictions offer large-scale projects; investors buy shares that yield carbon and biodiversity “dividends” without conferring land ownership. Market prices reveal demand and support liquidity. Compared to credit-based systems, this approach cuts costs, boosts credibility, and supports long-term commitments. It tackles the core problems holding back today’s markets—and offers a credible path to scale with real environmental impact.

Beyond World Literature (Lecture by Prof. Wiliam MARX)

イベント予定講演会/Lecture

Thursday, 8 May, 10:30–12:00 JST

World literature has been a reality since at least the nineteenth century. Texts travel across continents and cultures, translated from every language, taught in universities worldwide, and forming an emerging global canon. Never before have we been so free to read whatever we wish, from anywhere. Or so it seems. But is this true freedom, or merely a comforting illusion? What are the boundaries of this seemingly limitless literary exchange? This talk aims to explore those limits and propose a new approach to literature—a different way of reading texts, one that is either entirely new or, perhaps, simply old and forgotten. Welcome to the world library!

Previous Events

The worlds of the Black Death: new approaches (Lecture by Prof. Patrick BOUCHERON)

イベント予定講演会/Lecture

Wednesday, 16 April, 10:00–11:30 JST

The “Black Death” refers to the peak of the second plague pandemic, which spread across Europe from 1347. It remains the greatest demographic catastrophe in history. Today, interdisciplinary research—combining funerary archaeology, anthropology, microbiology, and environmental sciences—has transformed our approaches to it. Advances from DNA analysis to climate studies have contributed to a new understanding. Yet, the challenge remains: how to write a global history of a long-term event on a global scale? Though its precise geography is unclear, the plague’s routes trace the lines of force of connected worlds, mapping out a space that is discontinuous and global, like an archipelago.

What is the Purpose of Machines that Serve no Purpose? (Lecture by Prof. Dominique LESTEL)

イベント予定講演会/Lecture

Wednesday, 9 April, 2025 JST 15:00-16:30 JST

Despite the enthusiasm they generate, it is difficult to satisfactorily identify what humanoid robots could do that a human or a non-humanoid robot could not do better and more cheaply, and this observation constitutes a major challenge for the philosopher. Using the example of Sophia, the first humanoid robot to be granted citizenship in 2017, I will suggest that these strange machines occupy a sensitive place in our societies by taking on the role of messianic machines (machines that announce the coming of other machines), metaphysical machines (machines that force us to ask fundamental metaphysical questions such as knowing who is human or who is alive) and conjuring machines (machines that help to fight against the fear of dangerous machines).

Collaborations in Language: from Documentation to Resurgence (Lecture by Prof. Mark TURIN)

イベント予定講演会/Lecture

Friday, 4 April, 13:00–14:30 JST

In this richly-illustrated lecture, I discuss two collaborative partnerships in which I have been involved with historically marginalized, Indigenous communities in both the Himalayan region and in Native North America who are working to preserve and revitalize their languages. Through the presentation, I explore these three words: Collect, Protect, Connect.

Dealing with the Brussels Effect: How should Japanese companies prepare for the EU-AI Act? 3

イベント予定パネルディスカッション/Panel discussion講演会/Lecture

Wednesday, 19 March 2025, 12:00-13:00 JST

At the University of Tokyo, a webinar was held on December 11, 2024 and January 15, 2025, to explain the EU AI Act and the first draft of the CoP. In this webinar, we will provide an overview of the third draft released at the end of February and highlight important points that Japanese companies should particularly pay attention to.

The Role of Education and Science in the Digital Age (Yuval Noah HARARI)

イベント予定パネルディスカッション/Panel discussion共催/Joint Event

Monday, 17 March 2025, 3:00 - 4:30 pm (Doors open: 2:00 pm)

Today, digital networks provide us with an abundance of information. We invest more than ever in education and science. Despite these achievements our mental, socio-economic and political conditions have not improved. They seem to be even deteriorating. Why? What is going wrong? What can we do better? What can the first information revolution 600 years ago teach us? Two leading Japanese scholars in the field of media and AI governance will discuss these questions with Yuval Harari, the world-famous thinker and best-selling author, who explores the risks and opportunities of the information age in his new book Nexus.

Japanese as a Global Brand: Writing Japanese the European Way (Lecture by Prof. Viktoria ESCHBACH-SZABO)

イベント予定講演会/Lecture

Tuesday, 4 March 2024, 15:00-16:30 JST

This lecture explores the influence of the Japanese language on the global branding of Japan beyond its borders. It examines how the Japanese language is strategically employed to evoke distinct imagery, cultural significance, and authenticity. The session offers insights into the current landscape and future research directions of Japanese language as an important world language. Employing linguistic case studies from Germany and Hungary, the lecture highlights how Japanese writing elements are rephrased or combined with a product’s identity and with design cues evoking Japaneseness. Creatively adapted in new contexts overseas, the Japanese language has become a strong branding tool in Europe.


TOP