Video · Fireship ·
An unnamed OpenAI model just wrote an algorithm that solves 3-SAT faster than exponential time, breaking a 27-year-old hypothesis that over 800 published papers depend on. Let's break it down.
VideoFireship
Cursor shipped a backdoor, Polymarket got drained, North Korea pulled off the largest theft in human history, and Anthropic showed an unreleased Claude model can get root on any unpatched Linux server. Let's break it all down.
VideoFireship
The video examines whether recent AI infrastructure losses signal a bursting bubble, focusing on circular GPU financing, neo-cloud debt, and Anthropic’s proposed IPO. It argues that leveraged AI data-center businesses look vulnerable, while developers may still benefit from subsidized model access and falling compute costs.
VideoFireship
The video evaluates Kimi K3.1, a free 3-trillion-parameter open-weight model, against frontier systems on coding, cybersecurity, and 3D-generation tasks. It appears unusually strong on software benchmarks but is less capable on the practical 3D test, showing why benchmark harnesses and task-specific evaluation matter.
VideoFireship
The video covers a concentrated week of major internet-security failures, including PHP2Shell, vulnerable Node.js and GitLab releases, a backdoored AI package, weak certificate keys, rogue banking agents, and attacks targeting AI API credentials. It matters because AI-driven infrastructure expands the supply-chain and credential risks, making timely patching, dependency review, and agent monitoring essential.