The debates over copyright and AI training currently playing out in Congress, state legislatures, and, most notably, the courts, are focused on a few high-profile foundation model developers, but the outcomes of those debates will reverberate throughout the startup ecosystem.
While some startups are building foundation models themselves and will be directly impacted, the vast majority of startups are taking existing foundation models, fine-tuning them, and finding niche and industry-specific uses for AI. For instance, a startup might use multiple models to offer services that help local restaurants review supplier contracts or that help small waste hauling services identify business opportunities based on municipal construction permit approvals.
These uses are far removed from the original content that is included in the training data by foundation model developers, but the startups using AI in innovative ways in communities and industries across the country wouldn’t be able to build these tools without access to the underlying foundation models.
This report examines the downstream costs of licensing requirements for startups using LLMs specifically, but the takeaways are true regardless of the model and mode.
Read the report here.