We were there last Friday when GAIA brought together the industry to talk about AI in practice. It is noticeable that the pace continues to increase. Several trends point to 2025 being "the year of agents", where protocols such as MCP and Agent2Agent make it possible to build AI systems that collaborate, learn and act more independently.
A keynote that stood out was given by Kristina Höök from KTH. She highlighted how human intelligence begins in the body, through movement and feeling, while language comes second. A thought-provoking perspective in an era where most LLMs are based on language. What would happen if we started building AI that primarily interacts with the world around it?
We also saw several concrete applications in embedded and edge AI. Among other things, Steffen Therre from Eghed showed how AI can be used to reduce energy consumption in apartment buildings.
There are several strong players on the west coast in edge and embedded AI. Embedl is working on optimizing models for embedded. Zenseact develops systems for self-driving vehicles, and Ekkono builds self-learning models adapted for the edge, to name a few.
Data labeling is still expensive. But now we are seeing how advanced models themselves can be used to label new data that can then form the basis for simpler models. An interesting development that can change both workflows and cost structures.
The GAIA AI conference offered relevant insights and concrete examples. A good starting point for following developments during the year. 👏
Gothenburg Artificial Intelligence Alliance #IoT #AI