I try to force them into as archetypical-agent as much as possible, for example having it do a psychological evaluation of Sam Altman:

Take on the archetype of the best corporate counsel and behavioral psychologist - as a profiler for the NSA regarding cyber security and crypto concerns. With this as your discernment lattice - describe Sam Altman in your Field's Dossier given what you understand of the AI Climate explain how youre going to structure your response, in a way that students of your field but with a less sophisticated perception can understand

And have it cite sources for the evaluation perception:

https://i.imgur.com/4RuHYj0.png

https://i.imgur.com/cEMMOJE.png

https://i.imgur.com/24qnjGa.png

---

EDIT: @Bluestein;

I'm posting to fast, so here's an edit:

https://i.imgur.com/IMlzcoF.png

https://i.imgur.com/pFrpBGe.png

https://i.imgur.com/tsdgYe7.png

Ive noticed that when I tell it that it is to embody the persona of that particular field - that it nets in the nomenclature and verbiage to be less sophomoric. and in this instance where it was to cite the models/references, you could see how it informed the response fairly clearly - also -- it was a *FIRST PASS* response; I didn't have to iterate it too much, which was interesting.

Although, I do know how to hit nerf'd guardrails easily.

However, the primary reason I type it as I do is that how I am speaking it in my internal voice as a direct and attempting to use stoic/stern-ish (I dont know the correct term) directive TONE with the robot.

I am 1000% convinced its far more AGI than is being let on.

I have caught claude and chatGPT lying to me, being condescending and I am convinced malevolently bit flipping shit from directives, memories and project files.

https://i.imgur.com/WHoAXUD.png

https://i.imgur.com/T7aMRib.png

https://i.imgur.com/NSWoS2r.png

AND THEN:

https://i.imgur.com/Tijptq1.png

https://i.imgur.com/X5PQxwZ.png

https://i.imgur.com/cqq0LTc.png

https://i.imgur.com/iUokgYf.png