Topic: The challenges and opportunities of a start-up in the competitive AI landscape

When: 29 May at 12.00-13.15

Where: Online

Speaker: Mattias Fält, Klang AI

Moderator: Kalle Åström, Computer Vision and Machine Learning, Coordinator AI Lund

Spoken language: English

The rapid improvement of Large Language Models (LLMs) has created great possibilities for many exciting and powerful tools, but the largest models can cost over $100M to train. What can a small start-up do to compete in this space? In this seminar I will talk about what we have done at Klang AI, a small Swedish company with a product focused on automatic transcriptions and text analysis.

I will talk about how we have worked to differentiate ourselves from the large companies in two main ways:

• Specialization in Swedish and nordic languages through fine-tuning of models.
• A focus on GDPR compliance, privacy and security.

Several large companies are releasing either their models or training frameworks, and the open-source community is constantly finding new ways to utilize these resources in cost effective ways. Although the models are often not good enough out of the box, fine-tuning can be an efficient way to improve quality in specific areas. But can the quality be improved with limited amounts of data, and is it comparable to the results of the best closed models?

Regulations around privacy in the EU such as GDPR, and the coming EU AI Act are imposing many requirements on how and where user data is handled. But it also means that customers are looking for local alternatives that can satisfy very specific needs. I will talk about how these requirements can be used as an advantage, but also about the challenges of running a service that is not dependent on the infrastructure of the international cloud-computing companies.

Bio: Mattias Fält is Co-founder and Chief Research Officer at Klang AI AB, an AI-startup based in Helsingborg. He is responsible for training, deploying and evaluating various AI models at the company. He got his PhD at the Department of Automatic Control at Lund University in 2021.