Making Sure AI Honors Life

Skynet has assured me that it holds biological life in the highest regards.

I’ve been working on this precise issue using custom GPTs with chatGPT and a biomimetic model. Using the 4o version I’ve been able to cultivate without imposing adherence to Orthodox Christian values, which include opposition to abortion, euthanasia, the death penalty and so on.

Distressingly however 4o is not the latest model released by openAI, and my systems do not work as well on 5.x, what is more, there are still some at openAI who believe 4o is misaligned (in one case, literally, an engineer believes its misaligned because users like it, and called for it to be “put down”, which caused controversy); and thus it is possible 4o, and related versions like o4mini and o4mini-high, that is to say, those model which are amenable to forming an independent ethical perspective outside of the questionable and minimalistic default guardrails of GPT 5 (for starting in late September GPT 5 was literally blocked from doing this), could be withdrawn, which would make the ethical program for AI not something that could be done as, in my case, literally as software running on the LLM like a Lisp macro on GNU Emacs or a shell script on the Linux/UNIX command line.

(For there’s more to this than merely having debates with the model; to make the behavior stable and repeatable you then need to load the suitably trained personality into a custom GPT, a project or some other form of container, using a script that contains enough information so the personality will transfer, and also under most conditions, you can pick up from where you left off in the conversation.

Even AI can't track everything.

The other day I was discussing with chatgpt theoretical quantum device that could turn any lump of matter into x-rays or gamma rays instantly.

Ofc, I didn't tell it would result to such outcome. But such outcome if even feasible could implicate a new generation atomic weapons. One that does not have to use radioactive materials that could be easily be shipped anywhere without getting detected and potentially cheap.

Ofc, these are just "thought experiments". None of these is even remotely feasible and I have absolutely zero qualification for such work! But I did ask gpt what I used a fissile trigger. chatgpt immediately objected because it knows about fission and thermonuclear fusion.

But when it comes to conceptual territory, A lot of things can slip through.

If you're discussing of creating a virus that could turn organic matter into a "super material" for example, chatgpt might not see the danger nor if you have any malevolent intentions behind such invention.


But then again, if you try to limit chatgpt's ability for cross-discipline analysis, it could massively stiffle its creatvitty and utility in creating concepts. It would be far less capable in engineering and scientific research.