Looking at BIG-bench tasks and discovered the "canary string". A unique ID to prevent future large language models from training on benchmark tasks during web crawls.
Have we reached a tipping point for AI + culture loops? Or just the 2020's robots.txt?
