Take Back the media from the devil.

https://x.com/i/status/1951396858489864599

Scary

Yup!

Building in-house AI infrastructure for real-world tools like CAD, CNC, and robotic automation is exactly where local AI shines. Even better that you're approaching it from the manufacturing side — way fewer people are doing that than chasing chatbots.

Since you're planning to dive in for real, a few things to keep on your radar for when you're ready:

LLM Hosts: Look into Ollama, LM Studio, and [Text Generation WebUI] — good beginner-to-intermediate tools for running LLaMA/Mistral/etc. locally.

GPU Utilization: Your RTX 3060 is a capable 12GB card. Great for models up to 7B, and you can offload or quantize to go even further.

Open Source AI Toolkits: Check out:

ComfyUI (for AI image pipelines)

[AutoCAD LISP + Python bridges]

LangChain + [LLM Agents] for building tool-using AI systems

Router/SSL tips: For that whole SSL/router mess next time, you might want to:

Use Tailscale (for private networking)

Use Cloudflare Tunnel (to expose localhost without dealing with port forwarding)

Self-host reverse proxies like Caddy (auto HTTPS) or NGINX Proxy Manager

CAS latency (CL) refers to how many cycles it takes for RAM to respond to a read request.

In DDR5, 6000 MT/s CL30 is much faster than 6000 CL36 or CL40, because it responds in fewer clock cycles.

But there’s nuance:

True latency = (CL / Frequency) × 2000

So:

6000 MT/s CL30 → (30 ÷ 6000) × 2000 = 10 ns

6000 MT/s CL36 → 12 ns

5600 MT/s CL40 → ~14.3 ns

👉 CL30 at 6000 MT/s is ideal — you get tight timing and high bandwidth.

https://jonathanturley.org/2025/08/01/the-reveal-the-public-is-finally-learning-how-democrats-pulled-off-the-greatest-political-trick-in-history/