23 May 2025

Aeterna Labs Shares Expertise on Generative AI Implementation and Evaluation at Umeå AIMday

AIMday is a forum to bring together private and public sectors, where they can ask questions to researchers, allowing for the chance to bring science and evidence into their products and systems.

Photo by Lena Holmberg

The focus for this year’s AIMday, held on May 13th, 2025, was “Perspectives on AI,” a timely topic for organizations showing keen interest in leveraging generative AI. Aeterna Labs’ Researcher Anton joined to give his unique perspective; for the past few years, he’s been at the forefront of our projects at Aeterna Labs, using science to solve practical problems in contextual advertising. He provided some insights stemming from real-world application to the challenges organizations are facing in the rapidly evolving age of generative AI (GenAI).

“Everyone knows that they want to include generative AI. The issue is often to boil down brainstormed ideas to concrete minimum viable products (MVPs) that can be tested. Often, you run into problems with quality assurance, ultimately not trusting to put your brand in the hands of a generative AI’s output model. Since I’ve worked extensively with LLM evaluation on Aeterna Labs' real-world data, I was able to share some concrete suggestions on how to evaluate LLM output with the other workshop participants” said Anton. Of course, these evaluations are the necessary process to generate trust between businesses and AI.

Another heavily discussed topic was the ethics surrounding GenAI. With Umeå University being a prominent center for research in AI connected to Humanities and Society, the attending researchers offered valuable input for navigating potential pitfalls.

“It’s important to consider ethics from the beginning when implementing generative AI. Since there is no way to fully control the output of these models, you have to consider the stakeholders. As an example, are we forcing unwanted technology onto customers by replacing human support with a chatbot?”

There’s the element of timeliness, too. Anton says: “there is also a balance to strike between getting stuck on ethics issues too far ahead in the timeline and actually implementing an MVP that can provide more information to guide the next decision. From the discussions today, it's clear that Aeterna Labs is well-positioned in this area. We identify targeted applications where GenAI can make a significant improvement and make swift, small-scale implementations limited to specific parts of a system.”

In other words, testing internally is the name of the game, and keeping people at the center of these evaluations is what keeps the trust and accuracy in check. Anton notes: “This approach effectively lowers the overall risk while fostering innovation.” Indeed, a balance needs to be struck on the journey from ideation to implementation, and the steps between require “HI”, or human intelligence, to provide not just the best results, but results business owners can actually trust and consumers will not resent – like those frustrating customer service bots that never seem to quite get it.