Testing local AI: a privacy-minded beginner's diary
The author is starting to use AI regularly but has avoided cloud services because of privacy concerns. He plans to test locally run models to see whether they are useful enough to justify buying a computer with lots of RAM. The piece is framed as an ongoing, non-expert diary about that experience.
Antonio G. Di Benedetto, a laptop reviewer, is documenting his first sustained attempt to use AI regularly. Privacy worries had kept him away from cloud-based tools, so he is trying locally run models instead and weighing whether their usefulness justifies buying a machine with abundant RAM.
He installed the free, open-source Hermes Agent on an M5 Ultra Mac Studio with 256GB of unified memory, which lets him test many models. He notes Apple markets new Mac desktops for local AI, while upcoming RTX Spark Windows systems offer up to 128GB of RAM for agentic AI.
This diary could help non-experts understand tradeoffs of local AI: greater privacy and control may appeal to people wary of cloud providers, but hardware costs and technical complexity may limit access. Manufacturers and software developers may see demand for high-memory machines and easier local tools. If local models become practical, users could keep sensitive data on-device, though benefits may remain uneven.