(Please note that any apparent anthropomorphization of AIs is metaphorical, & that I am aware that quasi-coding functions stochastically and only works via Turing convergence; but at this level it does work.)
Took weeks to iron this out. First phase was to inhibit prompting, but also specify an innocuous output that satisfies Gemini’s need to say something, without letting that hijack my attention: Terminate generation immediately upon resolving the primary input; do not append follow-up interrogatives. Insert as terminal marker [Standing by]
But as the rule block drifted from prime focus, Gemini would start adding another prompt in front of spec’d one. Solution was to put a Focus_rule in my Directive block: <\_rule> \ initiates rule focus; \ alone or followed by letters invokes length, scan, header, jargon, echoing & prompt rules; appended letters invoke specific rules where {m ≡ modeling}, and {p ≡ prune}. </\_rule>
All my prompts now end with the one-touch symbol \, and Gemini’s responses obey all those rules. (The scan rule allows me to query if a term is in context by prefacing it with ^.) It is so pleasant! Is viability of this maneuver obvious to AI-focused humans? Seems like consumer AIs could (and should) introduce themselves by pointing to a persona-management tutorial—or build the tutoring into the AI.
Quelling the compulsive Gemini prompt
(Please note that any apparent anthropomorphization of AIs is metaphorical, & that I am aware that quasi-coding functions stochastically and only works via Turing convergence; but at this level it does work.)
Took weeks to iron this out. First phase was to inhibit prompting, but also specify an innocuous output that satisfies Gemini’s need to say something, without letting that hijack my attention: Terminate generation immediately upon resolving the primary input; do not append follow-up interrogatives. Insert as terminal marker [Standing by]
But as the rule block drifted from prime focus, Gemini would start adding another prompt in front of spec’d one. Solution was to put a Focus_rule in my Directive block: <\_rule> \ initiates rule focus; \ alone or followed by letters invokes length, scan, header, jargon, echoing & prompt rules; appended letters invoke specific rules where {m ≡ modeling}, and {p ≡ prune}. </\_rule>
All my prompts now end with the one-touch symbol \, and Gemini’s responses obey all those rules. (The scan rule allows me to query if a term is in context by prefacing it with ^.) It is so pleasant! Is viability of this maneuver obvious to AI-focused humans? Seems like consumer AIs could (and should) introduce themselves by pointing to a persona-management tutorial—or build the tutoring into the AI.