Video · Matthew Berman ·
More from Matthew Berman
All stories
VideoMatthew Berman
Deepseek crushed everyone
The video tests DeepSeek V4.1 Flash across 3D generation, games, multimodal vision, browser research, coding, and local app development. It argues that the model offers an unusually strong combination of quality, speed, low cost, and open-weight local deployment for AI builders.

VideoMatthew Berman
A Few Things We’re Working On

VideoMatthew Berman
Google "Carbon" just won everything...
The video reviews Google’s announced Gemini 4 Carbon, which Google claims leads major models in coding, science, agentic tasks, and cybersecurity while using fewer tokens and costing $2 per million input tokens and $10 per million output tokens initially. Carbon is not publicly available yet; Google plans a staged cybersecurity-partner rollout before broader developer, business, and consumer access, so the benchmark claims remain unverified in hands-on use.

VideoMatthew Berman
Who is local AI really for?
The video argues that local AI is currently expensive, difficult to configure, slower, and generally less capable than hosted frontier models, making it mainly suitable for technically skilled hobbyists. It explains that local AI's strongest benefits are data sovereignty, offline access, independence from AI labs, and owning intelligence, but predicts mainstream adoption will come through seamless hybrid routing between local and cloud models.