Running a frontier-class model on your own hardware is a procurement decision now: ~$2k for a small local model, ~$40k for near-Opus on four GPUs. Harder to find are the undocumented defaults…

• Hardware BOM for $2k (2x RTX 3090) or $40k (4x RTX PRO 6000) setups • Ready-to-run Docker configs for models like GLM-5.2-594B and whisper-large-v3 • PCIe Gen4 switch for peer-to-peer GPU…

1/10 ⭐4,357 T3MP3ST — autonomous red teaming platform. Multi-agent offensive-security meta-harness that stress-tests your AI systems before attackers do. 📰

5/10 ⭐1,482 claude-real-video — lets Claude (or any LLM) actually watch video. Scene-aware, deduplicated frames + transcript extraction. No more "I can't see videos". 📰

5/10 ⭐1,482 claude-real-video — lets Claude (or any LLM) actually watch video. Scene-aware, deduplicated frames + transcript extraction. No more "I can't see videos". 📰

5/10 ⭐1,482 claude-real-video — lets Claude (or any LLM) actually watch video. Scene-aware, deduplicated frames + transcript extraction. No more "I can't see videos". 📰

1/ GLM 5.2 and the coming AI margin collapse — how Chinese model pricing breaks the unit economics of Western AI vendors. 668 points on HN.

Dario and Altman giving you heartburn? One engineer published everything he knows about running LLMs at home. From a $2k build to a $40k one, with the config to match. • The exact hardware he runs •…

Local LLM setup guides are common. Ones that stay current with SOTA are not. Jamesob's GitHub guide covers running frontier models locally. 398 HN points suggests it fills a real gap.