Suggestions
Aaro Isosaari
Co-Founder, CEO at Flowrite
Aaro Isosaari is the Co-founder and CEO of Flow AI, a company specializing in AI evaluation and integration tools. Based in Helsinki, Finland, he has a strong educational background with a Bachelor's degree in Industrial Engineering and Management from Aalto University. His career showcases significant leadership and entrepreneurial experience across various roles.
Professional Experience
- Flow AI: As Co-founder and CEO since September 2020, Isosaari has been instrumental in developing products that help teams evaluate and improve AI applications. Flow AI originated from the earlier venture Flowrite, which focused on generative AI applications.
- Kiuas: He served as CEO from January 2019 to August 2020 and held various positions including Board Member and Program Lead until November 2021.
- Starting Up: Founder from April 2019 to July 2020, focusing on startup support.
- Slush: Worked in Investor Operations for six months in 2018.
- Nordic Sales Crew Oy: Led the sales team for Freska for nine months.
Contributions
Isosaari has been involved in several initiatives aimed at enhancing AI technology and its practical applications. His leadership at Flow AI has led to the introduction of models like Flow-Judge, which aids in evaluating large language model systems effectively.123
Online Presence
He is active on LinkedIn and Twitter, where he shares insights about AI developments and updates on Flow AI’s products.45
Highlights
The "Self-Taught Evaluators" paper by Meta AI introduces a novel evaluation approach for evaluating AI systems, reducing reliance on costly and unreliable human annotations.
Their approach uses synthetic data to train evaluators autonomously. The model iteratively improves its evaluation capabilities by generating and refining data.
Check out our recent blog post where we explore the paper's methodology, results, and implications.
w/ Bernardo Garcia del Rio / Flow AI
What is model merging and how is it different from model ensembling (in LLMs)?
Model ensembling involves combining the outputs of different models for a more accurate final output. It requires maintaining and deploying multiple models, along with a strategy to synthesize their outputs.
Model merging consolidates the weights of various models into a single model, simplifying inference as only one model is needed.
w/ Bernardo Garcia del RIo
