Scientists once hoarded pre-nuclear steel, and now we’re hoarding pre-AI content
1 min read
Summary
John Graham-Cumming, a former Cloudflare executive, has created a website called lowbackgroundsteel.ai to archive pre-AI, human-created content as a ‘time capsule’ of expression before AI-generated content became popular.
The name comes from atmospheric radiation contamination of steel after nuclear testing, which scientists remedied by using steel from pre-war shipwrecks, hence the name ‘low-background steel’.
The idea is to preserve pre-AI content to differentiate it from AI-generated content, which has contaminated research projects such as wordfreq, a Python library used to study language evolution across multiple languages, which had to shut down due to the rise of AI-generated text.
Some researchers are concerned that AI models may train on their own outputs, leading to lower-quality synthesis, but this can be avoided with real-world data.
