QH-E AI Statement
Executive Summary
QuadHelion will only use AI for coding tasks were it has previously established a hand-coded baseline for the programming task in question. AI output is only as good as it’s input and verification, therefore hand-coding as a baseline is a must, even if only once. The baseline entails all the normal books, tutorials, and docs hacking.
- QuadHelion shall join in the democratization of AI by engaging in a local first policy.
- QuadHelion will occasionally refrain from AI usage for two weeks or more to reinforce self-reliance, confidence, and spiritual re-alignment.
- Whenever AI can be avoided in favor of human cognition with acceptable overhead or valuable benefits to myself, my craft, and others, it will be.
Experience
Big Tech
QuadHelion has a 3 week lifetime experience with AI starting about the beginning of June 2026. This experience with online Big Tech AI by usage descending: Anthropic Sonnet 4.6 signed-in with a sideband email, VPN + Private Browsing Google Gemini, signed-in Grok, an hour or so of Mistral.
Local AI
Emphasis was placed on local AI usage with llama.cpp custom configured for either Vulkan or CUDA on a high-end consumer level nVidia card + NPU CPU. Local models used were:
- Qwen Coder Next 3.5
- Qwen 3 8B Medical
- Mistral 7B
- Gemma 4 E4B
- Google/Gemma 4 12B QAT Q4
Overall, even though these AI would often take many iterations with very precise and knowledgeable prompting to do medium-complex coding, other times they would one-shot the task with output beyond my own capabilities in seconds. That was indeed scary to witness as they were free versions. Instead of taking an hour to write a real nice Ruby or Bash sysadmin script, AI could write it perfectly well in seconds was the real gain for me as I write these scripts constantly. All the AI seemed to be most capable at Node JS apps, to which some of them were surprisingly good and robust.
Possible Future
- Prepare
- Plan
- Predict
It is quite possible that the Trillions of dollars of AI investment and AI Datacenters is the bubble that will not pop because these Datacenters are the actually planned end-game of “The Great American Panopticon” closely akin to the one nearly completed in China.
If you think these sadistic psychopathic Billionaires are stupid, my god are you the one that is epic stupid.
Therefore even though QuadHelion did use Big Tech AI, it did so experimentally and not professionally. I suggest to the freedom loving reader that you should in essence “know your enemy”, know what you are dealing with, and be ready to pull the plug the instant it should ever be turned on you.
The reader is highly encouraged to train a local AI for themselves.