A naukri.com initiative
Arstechnica
3w
8
Image Credit: Arstechnica
Scientists once hoarded pre-nuclear steel, and now we’re hoarding pre-AI content
- Former Cloudflare executive John Graham-Cumming has launched lowbackgroundsteel.ai, a website preserving pre-AI human-created content.
- The website aims to protect non-AI media from contamination by AI-generated content, similar to hoarding pre-nuclear steel during the Cold War.
- Generative AI models like ChatGPT have made it challenging to differentiate between human and AI-created content on the internet.
- The wordfreq project, tracking word frequency across languages, shut down due to the prevalence of AI-generated text online.
- Concerns exist about AI models training on their outputs, potentially leading to quality degradation known as 'model collapse.'
- Research suggests model collapse can be avoided by combining synthetic data with real data during training.
- Graham-Cumming aims to document human creativity from the pre-AI era through the website.
- The website points to major pre-AI content sources like Wikipedia dumps, Project Gutenberg, and GitHub's Arctic Code Vault.
- Lowbackgroundsteel.ai accepts submissions of pre-AI content sources to expand its archives.
- The project is a digital archaeology effort to distinguish between human-generated and hybrid human-AI cultures.
Read Full Article
Like
For uninterrupted reading, download the app