The article discusses the migration of 35kb preprompts from Opus to a self-hosted Ollama environment.

The author highlights concerns about privacy and data security with frontier providers like OpenAI and Anthropic, suggesting that running AI inference on third-party hardware risks exposure of sensitive information. The piece underscores the importance of operating on personal hardware to ensure verifiable privacy protections, as reliance on external providers may lead to unauthorized data usage and lack of transparency in their training pipelines.

Source: patrickmccanna.net