The World is Something We Make ...and we can make a truly open language model (www.argmin.net)

A recent talk titled “The World Is Something We Make” highlighted the urgent need for truly open language models in the evolving landscape of artificial intelligence. The speaker reflects on a significant shift in the AI/ML community, where the rise of large language models has led to a trend of consolidated and closed development practices, particularly due to concerns surrounding proprietary data usage. This regression is seen as detrimental, fostering an environment that hinders innovation and accessibility. The discussion draws from David Donoho's essay on "frictionless reproducibility," underscoring how open research—a practice that encourages collaboration and shared resources—has historically catalyzed breakthroughs in AI. To reclaim a culture of openness, the speaker advocates for a comprehensive approach that includes open weights, code, data, and processes. Initiatives like Marin and AI2 are mentioned as key efforts aiming for this level of transparency, but they face resource challenges. The call to action emphasizes the necessity of collaboration to enhance computing resources and leverage community participation in building open infrastructure. The potential for significant advancements is immense if a genuinely open model can be developed, leading to efficient training, fine-tuning, and auditing practices across the wider AI community.