Speaking on generative AI at LINE Yahoo's Osaka office
CosmoSketch representative Takino joined two of our LLM engineers to give a generative AI lightning talk at an event held at LINE Yahoo's Osaka office.
Introducing LLM-as-a-judge and its application
The talk focused on LLM-as-a-judge, a method of using large language models to evaluate text or AI outputs. It may automate part of the evaluation workload when people cannot review every output individually, and has attracted attention in generative AI product development and quality control.
We introduced recent research and discussed evaluation design for actual products and operations, using examples from our own validation work.
We also covered practical challenges: results can vary by model and instructions, and certain expressions or answers may be overrated.
We will continue assessing generative AI research and technical developments and applying the findings to sales support and operational systems.