Last November was a busy month for me. I had the chance to attend two major events back-to-back: the FCAI AI Day in Espoo and the Nordic AI Meet (NAIM) in Sweden. I gave the same talk at both events: “Detecting Latin in Historical Books with Large Language Models.” This represents the core of my PhD work during my first year. I also presented a poster at NAIM.
I want to give a big thank you to CASCADE for funding these trips. It was a great opportunity to step out of the paper archives and see what is happening in the wider AI world.
FCAI: Bridging Research and Industry
At FCAI in Espoo, the focus was heavily on connecting academia with industry. The Finnish Artificial Intelligence Society (FAIS) puts a lot of effort into this dialogue.
Of the projects presented that day, I was really impressed by the work from VTT. They are solving real-world industrial problems with highly innovative approaches. One project that stuck with me used semantic similarity to improve social science survey data. It was quite creative.
I was also excited to meet a research group working on historical data and multimodal models, a relatively rare combination! After my presentation, I chatted with people from a Helsinki innovation agency. They were interested in the potential commercial applications of our tech. We plan to stay in touch.
NAIM: Deep Dives into Interdisciplinary AI
I flew to Sweden on the 25th. The flight was fine, but the subsequent train ride to Norrköping was a bit bumpy. Despite the carriage movement, the journey went smoothly.
Norrköping is a fascinating place. It feels like a mix of old industrial history and modern life. The atmosphere is very calm and welcoming. The conference venue was especially cool. It was held at the Visualization Center C, combined with a nearby cinema. The organizers are really experts in picking such a creative spot.
At NAIM, the vibe was different. It felt more academic and interdisciplinary. One audience member of my talk asked about computational efficiency, which was a great question. It is exactly what I plan to address in the next steps of my pipeline design.
The keynote by Michael Felsberg from Linköping University was a highlight. His work on Computer Vision is truly world-class. His talk was so clear and well-structured, reminding me of my time at ICCV 2023. It was inspiring to see that level of rigor in Nordic.
Networking: From Sociology to German Beer
The best part of these trips is always the people.
I met many passionate scholars at the NAIM welcome reception on the 25th. Later, I attended many interesting presentations and viewed many posters. It was especially exciting to see how they use AI in fields like zoology and architecture. I particularly resonated with the sociological and educational perspective on AI, some aspects of which also originated from Helsinki.
On the evening of the 26th, I connected with another German scholar who is studying at Örebro University after dinner. I have always been interested in his Neural Symbolic approach, which has huge potential for structured data and complex reasoning. We went to Norrköping’s famous pub, Ölstugan Gull-Olle. Let’s just say I definitely felt the German passion and expertise for beer that night!
Reflections: Beyond the Hype
Overall, these events made me think about the state of AI. Right now, it feels like “LLM for everything.” But a lot of this work is still in the early exploration phase.
It appears that many projects lack robust verification. Factors like model selection and benchmark creation are often overlooked in favor of simply demonstrating functionality. In fields like ours, where LLMs can still hallucinate, rigorous testing is crucial. This makes our work on the Latin detection benchmark even more meaningful.
I appreciate the interdisciplinary and thoroughly ethical perspective from the Nordic AI community. Despite the relatively small population of these nations among European countries, I was impressed by the rigorous and open-minded atmosphere among scholars and their genuine concern for the human aspects in AI.
I hope we at CASCADE can continue to champion this kind of methodological rigor. As someone with an AI research background, I am always happy to contribute more insights on these methods. We should keep a close eye on frontier AI, but we must also ensure our progressive foundations are solid.
Yu Wu is a PhD researcher at the University of Helsinki, focusing on adapting computational methods to address large-scale challenges in historical research. He previously completed a master’s degree in computer science at ShanghaiTech University, where his research primarily centered on vision-language reasoning in AI and multimodal machine learning models. His current work draws from expertise in both natural language processing and computer vision, specifically targeting semantic matching and ensuring robust computational measures that can adapt to diverse types of historical data. Yu is passionate about using computational tools to bridge the gap between humanities and technology, with a long-term goal of contributing new methodologies that can advance historical and literary research.
CASCADE is a collaboration between University College Cork, University of Sheffield, University of Helsinki, KU Leuven, and Universität des Saarlandes. Funded by Horizon Europe under the Marie Skłodowska-Curie Actions (MSCA) Doctoral Networks and the UKRO.