“Helen Toner argues that if automated AI R&D significantly speeds up progress at closed labs, it would challenge the assumption that open-source models will always remain close behind in capability.”