Serkan Özcan
Posts
33 posts
MLX and RDMA: A Mac Studio cluster for large-model inference
Running large language models locally has often been defined by the memory limit of one powerful machine. The newer approach is to split a model across several computers and run them together over a fast interconnect.…
Read
Nemotron 3.5: Bringing agent intelligence from data centers to desktops
The important question in the agent market is no longer only which model is smarter. It is also where the model can run and under which data boundaries. The Nemotron 3.5 family approaches that question from both ends:…
Read
Muse Wonder: An agent needs boundaries as much as memory
The real value of an AI agent does not come only from producing a good answer. Following an insurance claim, understanding a long lease or building a working application requires keeping documents, tools and decision…
Read
RTX Spark: how I would evaluate a local AI computer
What interests me most about running AI on my own computer is being able to inspect the boundaries of the tools I use more directly. I would want to know which model is running, how much memory it uses and when it…
Read
Gemini agent: connecting work while keeping responsibility visible
After a meeting, much of the work can happen between applications. Finding notes, preparing a document and getting it to the right person look like separate small tasks. For me, the appeal of an agent that connects them…
Read
ChatGPT now chooses the shape of the answer too
When I ask an AI a question, getting an answer matters, but so does what I can do with it. A paragraph is sometimes enough. When comparing two options, I would rather see the same information side by side. I think the…
Read
How much of my Mac should an AI agent be allowed to access?
When I ask an AI assistant on my Mac to find a document, I feel that I know the boundaries of the task. But do the permissions I give the application match those boundaries? I think this will become one of the most…
Read
DigitalOcean MicroVMs: the cost of waiting for AI agents
When discussing the cost of AI agents, we usually focus on a model’s token prices. But a coding agent does more than generate answers: it installs packages, runs tests, and changes files. How we start and stop the…
Read
Anthropic’s distillation allegations: access and permission in AI competition
An AI model’s answer can become training material for another model. That possibility raises questions both about building more efficient systems and about how a company can protect its capabilities from competitors. The…
Read