About this blog

This site collects my research notes and write-ups on AI safety, with a particular focus on local AI and embedded AI — small, efficient models in the Gemma class that run on-device rather than in the cloud.

Running models locally changes the safety picture: there is no server-side moderation layer, updates ship slowly if at all, and the hardware budget constrains which safeguards are practical. The work published here explores what safety means under those constraints, alongside related topics in evaluation, alignment, and deployment of small models.

Questions or comments are welcome at arifsohaib.sa@gmail.com.