Global Science Network

YouTube channel covering hardware-based neural networks, offline AI, and DIY tech projects. Editorial position is privacy-first and skeptical of both big-tech and government control over information access. The author prefers reading pre-1975 textbooks for “less rewritten” sources — useful context when reading their takes on AI bias.

Channels

• YouTube: Global Science Network — hardware NNs, offline LLMs, DIY tech
• Site: gsnetwork.com (referenced for code/instructions in videos)

Content in This Wiki

• How To Run Private & Uncensored LLMs Offline | Dolphin Llama 3 — Walkthrough of running Dolphin Llama 3 from a $12 USB 3.0 flash drive via Ollama + AnythingLLM, fully offline. PowerShell-as-admin gotcha that loads the aligned base model instead.

Key Ideas

• Privacy-first stance: anything typed near a connected device is reachable by big tech and governments — local LLMs are the only real defense
• Recommends storing extra copies of uncensored model weights in a Faraday cage as a hedge against future regulatory bans
• Acknowledges that even an unaligned model still carries the bias of whatever lab trained the base — uncensoring removes refusals, not training-data bias
• Strong “garbage in, garbage out” framing for LLMs

See Also

• heretic — adjacent guardrail-removal tooling
• ollama, anything-llm — preferred local stack
• llama — Dolphin’s base