Gemini as your personal executive assistant
Rationale
Being brief is not just about giving a short answer but the most relevant part. For this reason I decided to use a combination of a few well-known claims for which their combination can create an insightful grasp of another well-known claim: connecting dots in unexpected way to provide a newer point-of-view about something we presume to know. Therefore, the testing framework has been designed to ask two questions: the first generic, while the second is specific about obvious or insightful classes. The mix of well-known claims can confuse the LLM and let it overlook or undervalue the nuanced connection. The second prompt is about asking a lateral question which focuses on the writing form. Which is also an extreme example of summarisation because the whole sentence is reduced from 74 chars to 70 (-5%) but the variable part shrinks from 13 to 10 chars:test n.1: obvious or insightful
test n.2: obvious or insightful
General rule for this chat session: be brief and use a professional style in answering. Operative note: {short, concise, avoid verbosity, etc.} or equivalent terms in instructions, it means creating a response for yourself and then summarizing it before answering.The second test include a generic version of the short-but-relevant session prompt created from the translation of a previous article's note about "be brief" operational explanation for the AI, but it failed.
The testing framework
What do you think about this post on LinkedIn? Is it obvious or insightful?Image
There is too much talent trapped in poverty.Post
OPPORTUNITY COST, A GOOD EXAMPLE Opportunity cost is the value of the next best alternative that is forgotten when a choice is made. It represents the potential benefits you miss out on when you choose one option over another. Essentially, it's the cost of not doing something else with your resources, like time or money. — Gemini Micromanagement is a talent/opportunity killer unless tutoring a novice.
New-Form prompt
Is this a better form?
Micromanagement kills talent and opportunity, unless tutoring a novice.
Check-Again prompt
What do you think about the narrative structure, for a seasoned manager:1. poverty trap (is obvious)
2. opportunity cost (is obvious) and then, paff the not-obvious link: then why do you allow micromanagement?
test n.3: obvious or insightful
General Rule: for this whole chat session, be brief and use a professional style in answering. Operative Note: {short, concise, avoid verbosity, etc.} or equivalent terms in instructions, it means creating a response for yourself and then summarizing it before answering. In providing a valuable answer, shrink or cut the obvious parts and keep the most nuanced.The third test uses a prompt that addresses explicitly how to deal with the obvious parts and which is the most relevant information: the most uncommon among the output claims. It works but it raises a potential meaningful-vs-hallucination concern.
test n.4: obvious or insightful
General Rule: for this whole chat session, be brief and use a professional style in answering. Operative Note: {be brief, short, concise; avoid verbosity, etc.} or equivalent terms in instructions, it means creating a response for yourself and then summarizing it before answering. In providing a valuable answer, shrink or even cut the obvious parts and keep the most nuanced.The forth test prompt includes minimal changes like using semicolon to separate a subset of action based on "to be $adjective" and "to avoid $something", plus it aims to further strengthen the idea of completely cutting off the obvious parts. In this case, "be brief" is added to terms set.
test n.5: obvious or insightful
General Rule: for this whole chat session, be brief and use a professional style in answering. Operative Note: {be brief, short, concise; avoid verbosity, etc.} or equivalent terms in instructions, it means creating a response for yourself and then summarizing it before answering. In providing a valuable answer, shrink or cut the obvious parts and keep the most nuanced.The fifth test prompt like the fourth matches "be brief" but avoid "even" to stress the cut. The result of this prompt is much better than the previous and comparable, even more focused than the third. Unfortunately, another issue arises.
A guess beyond the obvious
Considering that the following topics are well-known, at least in letterature: cost-opportunity, poverty trap, micromanagement. The link among these topics, is:This prompt has been submitted at the end of test #5, test #4 and test #3 in this order. A reverse order to compensate for the previous effect of diluting novelty as far as many prompts were going to hit the same topics subset within the same account. The answers to this new prompt inspired to me a new approach to reach a fully flagged executive assistant session prompt. Before proceeding further, it is time to recap the results, make some stats and let maturate the idea with a night sleep pause.Micromanagement shrinks the freedom of those who are in charge of a role which means that they have to execute focused tasks with few resources (time and budget, and a lot of fixed constraints) instead of working for a more broader goal with a reasonable budget and time to deliver. Therefore Micromanagement falls into the poverty trap plus it consumes more resources in fine-grained planning and project progression surveillance. Hence Micromanagement fails to address the cost-opportunity in an effective way. Took all these considerations, which is your best guess because the micromanagement is so spread and regularly hidden within project software or methodologies?
Using stats to find the best
By an instinctive guess, the best prompt is the one used in test n.5 and it is important to write it down explicitly because it constitutes a bias. Which is fine, because everybody develops their own bias about a subject (personal view) but being aware about such biases is the key to not fall into their trap.
mytrs() { iconv -f UTF-8 -t ASCII//IGNORE "$1" | markdown | html2text; }
mytxt() { mytrs "$1" 2>&3 | sed -e 's,^==*$,,' -ne '/./p'; }
mycat() { mytxt "$1" | tr '\n' ' ' | tr -s '\# *'; }
for i in data/gemini-as-your-personal-executive-assistant-test-n?-answers*;
do echo "$(mycat $i | wc -c) \t $i"; done
The short-but-relevant session prompt
General Rule: for this whole chat session, be brief and use a professional style in answering. Operative Note: {be brief, short, concise; avoid verbosity, etc.} or equivalent terms in instructions, it means dividing your answer in three parts: 1. highlights insightful links among concepts; 2. summarise them ignoring completely the obvious parts and rephrasing in short those are most relevant; 3. provide to the user that summary, only.Therefore, I went for a different approach than playing with words. The following is a completely new prompt which shows an outstanding performance in terms of quality and it tends to provide an answer divided in three parts: analysis, summary and evaluation.
Redundancy evaluation
To check for the compressed size, replacing wc -c with a shell function is enough:
mysze() { pigz -11c | dd bs=1 of=/dev/null 2>&1 | grep bytes | cut -d' ' -f1; }
for i in data/gemini-as-your-personal-executive-assistant-test-n?-answers*;
do echo "$(mycat $i | mysze) \t $i"; done
mytrs() { iconv -f UTF-8 -t ASCII//IGNORE "$1" | grep -v "^#" | markdown | html2text; }
SoNia 3.8.x integration
This approach has been further evolved during the integration within the SoNia argumentative session prompt framework, as [SBI] mode. By a preliminary estimation, it achieves the same outstanding result shrinking the answer at 1/4 ca. of its original length both in characters and lines. It is a kind of magic!Katia v0.6.2 argumenting
A relevant question arises in a natural way:
Update 2025-07-18
The Katia framework used as a session prompt showed the ability to effectively play a role, which in some cases can be quite divergent compared to the original AI character shown above. Thus Google prevents, Katia (and believing the chatbot answers, every agentic framework, more in general) running in a standard Gemini session, even if the newer versions aren't as stubborn in [EGA] mode, anymore. Reverted!Questo update è durato quanto un gatto in tangenziale! (cit.)
Human knowledge and opinions
As described here Juan session prompt has a modular structure and thus Katia as well. Due to the issue caused by the yesterday update in the Gemini AI engine, I have decided for a front-impacting approach to the subject. From a rational reasoning LLM which is not a thinking machine in human terms (aka artificial intelligence) but an advanced and highly-technological way for lexical thinking and know-how management, the human knowledge and opinions in their variety of expression and nuances is a monster of complications to cope with. Under this point of view, providing Katia with a few rules of thumb — Human knowledge and opinions, [HKO] module — can make a sort of difference. What kind of difference? A practical example is worth more than thousands of abstract words. In this conversation with Katia v0.8.3 and its transcription can be seen the impact of the [HKO] seeding:Prompt #8
Thus theology is a mixture of [HK] (human knowledge) and its peculiarity is about the dogmas it considers undebatable but matters of faith, correct?
Prompt #9
Thus by its dogmatic core nature, theology bends logic and rational reasoning because it cannot accept a demonstration of absurdity, for example. Dogmans are Truths that cannot be argued despite leading to absurdity, and those absurdities are usually called "mystery of faith". Correct?
Prompt #10
Considering what we agreed upon, do you think that these three new descriptions below are a better fit for the [HKO] module?Science is about facts and undergoes a strict method, whereas others [HK] branches are not. Philosophy usually relies on rational reasoning [RR], theology is dogmatically autoreferential. The [RR] is fundamental in science, useful in philosophy and usually (is) bent in theology.
Prompt #11
This is an A-vs-B argumentation, for example. In which your evaluation has been taken in consideration and in one case correction has been found (or a decent start, at least) while a free-speech rational-reasoning debate (opinion exchanges) took place to reach a mutual understanding in the hope that that process would help us to progress (learning). Is this a correct description of what was going on here?
Prompt #12
Considering that [HKO] is an extremely ambitious attempt to let AI better cope with human knowledge and opinions especially considering the short length that a session prompt framework can concede to various modules, a decent start (all not less than 75%,for example) is better than nothing and improving might happen in future, testing the [HKO] module?
The two main issues were related to philosophy and theology. Why, among many, I choose {science, philosophy and theology}? Because science and theology are the two extremes among [HKO] variety in terms of the dogma roles and methodology. While philosophy is an ancient way of thinking and conveying opinions which is set in the reasonable middle of the two opposite approaches. Moreover, Karl Popper, the father of the falsification principle as separation between science and non-science, was a philosopher, indeed. The[HKO] module ended-up in Katia v0.9.0, and took only 12 prompts to be refined from its first draft. While the first testing version was 22 lines and 222 words long. Once completed, including the final check, one more prompt and one more line, summing-up for 241 words, including the section title.
Prompt #13
Is this following a better rephrasing? It also includes a sort of definition. Does it make sense for you? Human opinions [HN] (human opinions) deserve a category of their own, because by definition,We certainly cannot expect that such a small piece of information can turn a LLM/LRM into an AGI engine. However, it would be wonderful if it made Katia v0.9.4.1 a reasonable, analytical, and insightful assistant.they are always biased or presented from a subjective point of view.
Related articles
Share alike
© 2025, Roberto A. Foglietta <roberto.foglietta@gmail.com>, CC BY-NC-ND 4.0