- The Luddites are back, but this time as technologists opposing AI progress through the 'Poison Fountain' project.
- Poison Fountain aims to contaminate internet data used by AI systems to slow down their development.
- Geoffrey Hinton, the 'godfather of AI,' warns of existential dangers posed by advanced AI.
- Historical examples show violent backlashes against disruptive technologies, like the Luddites and attacks on 5G towers.
- Large language models (LLMs) train on internet data, making them vulnerable to poisoned content.
- Poison Fountain's strategy involves embedding links to flawed data to degrade AI models during training.
- Recent research suggests even small amounts of poisoned data can significantly harm LLM performance.
- Challenges for Poison Fountain include AI companies' data cleaning processes and the vastness of the internet.
- The project highlights a structural vulnerability in LLMs due to reliance on untrusted, scraped data.
- Poison Fountain signals the start of a cat-and-mouse game as AI becomes more embedded in daily life.