Part 7/15:
Miikkulainen emphasizes the need for open-source foundational models in AI research. Open models, like LLaMA, enable scientists worldwide to analyze, understand, and improve AI systems, fostering a vibrant community that can address issues like bias and safety. He advocates for shared infrastructure and resources that democratize AI development, ensuring that progress is not limited to corporate entities with proprietary models.
He notes that currently, building massive foundation models demands significant hardware and financial resources—barriers that restrict research and transparency. Initiatives to pool academic and public resources can help overcome these challenges, leading to models that are more accountable and equitable.