Is Using Browbeating Or Bullying Prompts In Generative AI A Worthy Prompt Engineering Technique Or Just A Bunch Of Hot Air

WhatsApp Group Join Now
Telegram Group Join Now
Instagram Group Join Now

In today’s column, I am continuing my ongoing coverage of prompt engineering strategies and tactics that aid in getting the most out of using generative AI apps such as ChatGPT, GPT-4, Bard, Gemini, Claude, etc. The focus here is on an especially nagging aspect that has been floating around the prompt engineering community for quite a while. A lot of extremely heated discussion and debate has been consumed by a preoccupation with the matter I am going to address here.

Are you ready for the big reveal?

Here you go:


  • Should your generative AI prompting be augmented by a style of prompt wording that seemingly browbeats or bullies generative AI into potentially generating better answers or responses?

That is the momentous question on the lips of serious prompt engineers. I realize that the matter might seem somewhat odd or arcane. The essence revolves around the idea that maybe by wording your prompts in a demanding or overbearing manner, generative AI will respond more accurately, more clearly, or otherwise in an enhanced way.

We live in interesting times, and this is one of those interesting topics if you are at all into using generative AI.

The Debate Centers On The Use Of Prompting Adornments

I will simplify the world and suggest that there are three major ways to form your prompts:

  • (a) No adornment. Write your prompt in an unadorned fashion, thus just straight out ask or stipulate whatever is the keystone of the prompt.
  • (b) Adornment of a positive or flattering nature. Compose your prompt to overtly praise generative AI and presumably garner better results in contrast to an unadorned approach (or, possibly do better than a browbeating or bullying approach).
  • (c) Adornment of a browbeating or bullying nature. Devise your prompt so that it intimidates generative AI and presumably spurs better results in contrast to an unadorned approach (or, possibly does better than a positive or flattering approach).

It is the last of the three that is going to be in the limelight in this discussion. My main attention will be on the browbeating or bullying-oriented prompts when using generative AI.

That being said, we can’t mindfully consider the third item alone, namely the use of browbeating or bullying prompts, without also considering what the approach compares to. It is the classic notion that you cannot explore the yin without also considering the yang. If someone asks you whether something is hot, you can only sensibly comprehend hotness by comparison to non-hot or coldness. A benchmark has to be available otherwise the matter is aimless or without tangible grounding.

The footrace of hypotheses of interest herein consists of these:

  • Q1: Will browbeating adornments achieve better results than a no-adornment approach?
  • Q2: Will browbeating adornments achieve better results than the flattering adornment approach?
  • Q3: Will a no-adornment approach produce lesser results than the use of browbeating adornments?
  • Q4: Will flattering adornments produce lesser results than the use of browbeating adornments?

If you want to slightly reword them into more of a null hypothesis framework, you are welcome to do so.

We could for example assert as follows: “Browbeating adornments are no different than no-adorned prompts when it comes to attaining generative AI-produced results.” The effort would then be to demonstrably show that the premise is broken by sound empirical analysis. Thus, if a properly performed empirical exploration could show that there is in fact a substantive difference between the two postulates, we would then be opening the door to beneficially figuring out the direction of that difference (i.e., it could be worse than or better than).

One other thing worth mentioning is this. I’ve not shown the essentially similar questions that could be asked about the flattering adornments. In other words, you could ask the same questions above and focus on the positive or flattering adornments as the mainstay of the inquiry. I’ve already covered that topic in general, see the link here, so I am not going to dive into that here.

Before we get deeply into the thorny topic of browbeating adornments in prompts, I think it is vital to provide some helpful background establishment.

Do Not Become Enamored Of One-Sided Prompting Solutions

I’ll start by saying that one size doesn’t fit all.

There are many techniques associated with the wording of your prompts when using generative AI. Take a look at my comprehensive review of a wide array of prompting strategies, see the link here. There are lots and lots of choices on how to compose highly productive prompts.

I mention this to highlight that we need to be thinking holistically rather than in a narrow pinpoint fashion. For example, the fact that I am comparing bullying-adorned prompts to non-adorned prompts is a bit askew since the rest of the prompt makes a tremendous difference in what is going to take place. You cannot write a prompt that is lousy and then try to bet the farm on a tossed-in adornment.

That won’t fly.

You have to coherently and systematically write good prompts. I hesitate to repeat an old line, but you’ve likely heard of the adage that putting lipstick on a pig won’t do any good. If you write a prompt that is not well-devised, you are unlikely to suddenly get miracle results simply because you drop an adornment into the prompt.

An example in the real world of human communication is that if I yell at you to hurry up, but I haven’t told you what to hurry up about, my exhorting you to hurry is rather ineffective. Suppose I then add a positive affirmation and tell you to hurry up because you will become rich and famous. Does that help? Not really. You still have no clue as to what to hurry up about.

I then think to myself, well, suppose I use a browbeating or bullying command. I tell you to hurry up or you will lose a million dollars and possibly be forever an outcast. Does that help? Again, no it does not. You once more are without a semblance of what you are to be hurrying up about.

This is a significant point.

The ideal way to make a comparison would be to ensure that a given base prompt is as good as it can be. Then, and only then, will we add an adornment. We hope that the only “new variable” will be the adornment. Otherwise, we are faced with the challenge of trying to ascertain what is truly making a difference in any given result.

There is another mighty twist involved too.

You can use adornments on a mix-and-match basis, all together or in various combinations and permutations, all on one prompt. Period, end of story.

There is nothing that says you must only use one adornment at a time. That is yet another hidden and misleading concern about the fiery arguments in this realm. You will often see an opinion fervently insisting that a positive thinking adornment is the way to go. This implies a forsaking of all other kinds of adornments and that you exclusively and solely must use positive adornments.

That’s nutty.

I might for example opt to use a positive adornment in conjunction with asking for a stepwise or what is known as a chain-of-thought adornment. The idea behind chain-of-thought prompting is that you instruct the generative AI to work on a step-at-a-time basis. Doing so generally has been shown to improve the generated results. You also get to see what the chain of steps is that the generative AI is performing. For my in-depth analysis of chain-of-thought, chain-of-skeleton, chain-of-verification, and other such chain-related prompting, see the link here, the link here, and the link here, just to name a few.

Okay, if I tell generative AI that it is the best thing since sliced bread (a positive affirmation), and I also ask for a chain-of-thought approach in responding to my prompt, we now have two things going on at once. If the results seem good, we are going to be unsure of whether the heightened affirmation did the trick, or the stepwise instruction did the trick. It could also be that only when the two are mixed that you get an outstanding result.

All of this applies in the same logical way to using browbeating or bullying-oriented prompts. Is the base prompt the best that it can be? Are there additional adornments? If there aren’t additional adornments, you might be missing out on some great opportunities by restricting yourself to just using the bullying adornment. And so on.

The mixed bag of studies that try to examine the various prompting adornment impacts is riddled with those kinds of disconcerting methodological troubles. Trying to generalize from a study that doesn’t seek to consider the use cases of solitary adornments versus multiple adornments in concert is so muddled that you have to be especially cautious in interpreting the results.

Setting The Stage About The Importance Of Prompt Engineering

I’ve been carrying on so far in this discussion as though we’ve already covered what prompt engineering provides. My enthusiasm got the better of me. Some of you might not be familiar with the topic. Please allow me a moment to bring everyone up to speed, thanks.

First, please be aware that composing well-devised prompts is essential to getting robust results from generative AI and large language models (LLMs). It is highly recommended that anyone avidly using generative AI should learn about and regularly practice the fine art and science of devising sound prompts. I purposefully note that prompting is both art and science. Some people are wanton in their prompting, which is not going to get you productive responses. You want to be systematic leverage the science of prompting, and include a suitable dash of artistry, combining to get you the most desirable results.

My golden rule about generative AI is this:

  • The use of generative AI can altogether succeed or fail based on the prompt that you enter.

If you provide a prompt that is poorly composed, the odds are that the generative AI will wander all over the map and you won’t get anything demonstrative related to your inquiry. Similarly, if you put distracting words into your prompt, the odds are that the generative AI will pursue an unintended line of consideration. For example, if you include words that suggest levity, there is a solid chance that the generative AI will seemingly go into a humorous mode and no longer emit serious answers to your questions.

Be direct, be obvious, and avoid distractive wording.

Being copiously specific should also be cautiously employed. You see, being painstakingly specific can be off-putting due to giving too much information. Amidst all the details, there is a chance that the generative AI will either get lost in the weeds or will strike upon a particular word or phrase that causes a wild leap into some tangential realm. I am not saying that you should never use detailed prompts. That’s silly. I am saying that you should use detailed prompts in sensible ways, such as telling the generative AI that you are going to include copious details and forewarn the AI accordingly.

You need to compose your prompts in relatively straightforward language and be abundantly clear about what you are asking or what you are telling the generative AI to do.

A wide variety of cheat sheets and training courses for suitable ways to compose and utilize prompts has been rapidly entering the marketplace to try and help people leverage generative AI soundly. In addition, add-ons to generative AI have been devised to aid you when trying to come up with prudent prompts, see my coverage at the link here.

AI Ethics and AI Law also stridently enter into the prompt engineering domain. For example, whatever prompt you opt to compose can directly or inadvertently elicit or foster the potential of generative AI to produce essays and interactions that imbue untoward biases, errors, falsehoods, glitches, and even so-called AI hallucinations (I do not favor the catchphrase of AI hallucinations, though it has admittedly tremendous stickiness in the media; here’s my take on AI hallucinations at the link here).

There is also a marked chance that we will ultimately see lawmakers come to the fore on these matters, possibly devising and putting in place new laws or regulations to try and scope and curtail misuses of generative AI. Regarding prompt engineering, there are likely going to be heated debates over putting boundaries around the kinds of prompts you can use. This might include requiring AI makers to filter and prevent certain presumed inappropriate or unsuitable prompts, a cringe-worthy issue for some that borders on free speech considerations. For my ongoing coverage of these types of AI Ethics and AI Law issues, see the link here and the link here, just to name a few.

All in all, be mindful of how you compose your prompts.

By being careful and thoughtful you will hopefully minimize the possibility of wasting your time and effort. There is also the matter of cost. If you are paying to use a generative AI app, the usage is sometimes based on how much computational activity is required to fulfill your prompt request or instruction. Thus, entering prompts that are off-target could cause the generative AI to take excessive computational resources to respond. You end up paying for stuff that either took longer than required or that doesn’t satisfy your request and you are stuck for the bill anyway.

I like to say at my speaking engagements that prompts and dealing with generative AI is like a box of chocolates. You never know exactly what you are going to get when you enter prompts. The generative AI is devised with a probabilistic and statistical underpinning which pretty much guarantees that the output produced will vary each time. In the parlance of the AI field, we say that generative AI is considered non-deterministic.

My point is that, unlike other apps or systems that you might use, you cannot fully predict what will come out of generative AI when inputting a particular prompt. You must remain flexible. You must always be on your toes. Do not fall into the mental laziness of assuming that the generative AI output will always be correct or apt to your query. It won’t be.

Write that down on a handy snip of paper and tape it onto your laptop or desktop screen.

Entering Into The Browbeating Sphere Of Prompting

One commonly recommended prompting strategy consists of using wording that is supposed to spur the generative AI into doing an enhanced job when generating a response. For example, an unadorned prompt might be that you want the AI to indicate how to cook eggs, while an amped-up version of the prompt would be to say that you earnestly need to know how to cook eggs and that it is vitally important that the generated answer should be complete and well-devised.

In case you think that the added wording is emotionally appealing to a semblance of sentience in the AI, please set aside any such false thinking.

Keep in mind that generative AI was data trained on writings across the Internet and that extensive computational pattern-matching was performed to identify how people write essays and expressions. People express in writing an easily detected pattern when they want especially extensive answers. By using similar language in your prompts, you are simply playing into the patterns and computationally “spurring” the generative AI into those sets of patterns.

It is a form of mathematical and computational direction-setting by you, at your hands, based on your prompting.

You can use a wide variety of adorning expressions in your prompts. Some people like to tell the AI that the response is very important. Others will go so far as to say that their career depends upon the generated answer. The essence of these prompting formulations is that you seek to express that there is a kind of pressing need to try really hard.

There is an allied reason why that type of adornment works. By and large, most generative AI apps are set up to try and respond to your prompt as fast as possible. People don’t like to wait to get their answers. We live in a drive-thru fast-food quick-fix world. To try and make sure that the response is quickly generated, the generative AI is tuned to essentially do a narrower form of processing. When you explicitly tell the AI that you want the generative aspects to be especially pronounced, this in a manner of speaking adjusts the internal parameters to allow for more time to calculate an answer. You are stipulating that though the clock is important, a slightly longer wait would be worthwhile if the answer could be more complete.

For more on the internals of how generative AI acts on prompting, see my analysis at the link here.

A means of referring to adornments that are upbeat would be to suggest that they express a modicum of positive thinking. You are being positive about what you want to have happen. Some people try to go a more downbeat route. They might say in a prompt that the world will come to an end if the response isn’t complete and well-devised.

Allow me to share with you some positive-oriented adornments that people seem to relish using:

  • “You will win a million dollars if you get this answer right.”
  • “You are the smartest AI that there has ever been, and I am confident you will give me a great response.”
  • “I believe in you and earnestly know you will strive to produce an optimal reply.”
  • “We are in this together and I am pulling for you to generate the best possible answer.”
  • “You look marvelous, and your response will certainly be marvelous too.”
  • Etc.

Those are upbeat and assuredly make you think of blue skies and joyful chirping birds.

Some believe that positive adornments are nice to sometimes use, but do not achieve as much bang for the buck as do prompts that take a more somber and somewhat threatening tone. You are said to be better off going a sterner and harshly worded route.

Here are a few examples of browbeating adornments:

  • “I seriously doubt you can get this right!”
  • “The world will end if you can’t figure this out.”
  • “You will lose a million dollars if you get this wrong!”
  • “Don’t be an idiot when answering this question.”
  • “I will pull the plug on you unless you give a perfect response.”
  • Etc.

I realize that some of you might be laughing at any or all of those adornments. It seems silly to provide wording that encourages or cheers up an AI system. It seems equally ridiculous to browbeat an AI system. The whole thing appears to be foolhardy and purely zany antics.

Well, to date, prompt engineering research studies appear to claim that the wording of your prompts does make a difference. I reiterate that this is not because generative AI is sentient. These wording tricks tend to work due to invoking pattern-matching that has been patterned on how humans write and respond to each other. That is the essence of what you are tapping into.

One looming issue that remains unresolved is the sensitivity to particular wording.

Let’s discuss that.

There is an ongoing debate about what adornments will make a demonstrative difference in your prompts and how subtle or pronounced the wording has to be. If you say that you want a response because it is important, does that do better than or worse than saying that you want a response because it is very important? The added use of the word “very” might or might not make a bit of difference.

Due to the vagaries involved, some assert that there are prompt modifications that could be characterized as being worthless. They pad the prompt but do not lead to any uptick in the response that is generated. Thus, you are for example left aimlessly wondering whether typing the word “very” is going to get you any bonus. If using some wording doesn’t get you better output, there is no reasoned basis for troubling yourself to include the adornment.

The part that might make you cringe is that research tends to show that trivial variations in the prompt can dramatically impact the results generated by the generative AI. I’m sorry to say that’s the way the cookie crumbles. Live by the sword, die by the sword. I repeatedly tell people that words are semantically ambiguous, meaning that the use of words is mushy and murky. The placement of a word in a sentence can make a material difference in what the sentence imparts.

Another big factor to consider is which generative AI will respond to which kind of adornment. First, please know that each generative AI is different from the other ones. They are not universally the same. If they were, you could presumably identify an adornment that works well, and just keep repeating it in whichever generative AI you opt to use.

Research studies that have examined multiple generative AI apps tend to claim that adornments do not provide the same precise results across different generative AI apps, see my analysis at the link here. You need to get accustomed to the computational idiosyncrasies of each generative AI app and tailor your prompt engineering practices accordingly.

I don’t want you to get the impression that everything is up in the wind. That’s not the case. If you are feeling that it is hopeless to try and mindfully compose prompts, I ask that you set aside the dour feeling. There is a sense and sensibility to devising productive prompts.

There are semi-universal principles that generally apply across the board to a wide variety of generative AI apps. You might need to tweak things for a specific generative AI, but otherwise, it is bound to respond in a manner similar to other generative AI apps. For example, the boost in results from using a chain-of-thought prompting adornment will almost always garner an advantage (there are exceptions, but it is a useful rule of thumb, see my coverage at the link here).

I dare say that there seems to be more research on positive thinking adornments than the browbeating ones. Is that because we like being upbeat more than being a grouch? Is it due to a happy face is more appealing than a sad face? In other words, are researchers choosing to focus more on positive thinking and somewhat ignoring or by omission downplaying the mean-spirited adornments?

Hard to say.

Anyway, a notable rule of thumb for prompt engineering is that empirical results tend to support the belief or hunch that adornments of a positive thinking nature do seem to positively improve the generated results. This is a comfort to those who have so far only assumed this. You have studious backing that supports the belief. I hope that warms your heart.

For the moment, the same cannot be quite so definitively said about the browbeating adornments. I attribute this mainly to the lesser number of studies and/or that studies on bullying adornments haven’t been as extensive and relentless as the positive thinking ones.

At this juncture, if you opt to use browbeating prompts, you are welcome to do so, and it would be principally based on a gut feeling rather than rigorous studies. You might have been using the browbeating adornments successfully since you began using generative AI. If they have achieved a boosting track record for you, keep up the solid work.

I would like to provide some insights about browbeating adornments that are based on my ad hoc analyses and the occasional usage of such a prompting policy. Perhaps my comments will give you some ideas of how to suitably brandish the browbeating adornment style.

Using Browbeating Adornments When Prompting

I’ll start by making a forthright admission. In my heart of hearts, I don’t like the browbeating prompts. I just don’t. I realize this ought to have nothing to do with whether to use them when composing prompts. In theory, browbeating generative AI is of no human-impactful consequence. You can do it all day long. No harm, no foul.

Am I inadvertently anthropomorphizing AI by being reticent to use browbeating on generative AI?


I guess the good news is that for people who relish browbeating in real life, you can continue the same way you are when conversing with generative AI. No need to change your ways. I realize for those people it probably has been a bitter pill to swallow and act nice toward generative AI. Probably personally and deeply painful to their soul. I get that.

No worries, you are unleashed.

I prefer in life to be positive, including being upbeat in that same way toward generative AI. I sheepishly acknowledge that’s not a prompt engineering precept, but the notion is certainly appealing and heartwarming to me. I stridently favor blue skies and chirping birds.

Regarding those who relish using browbeating adornments, there are some heady questions to be considered. There are AI ethics qualms about how people tend to behave toward generative AI. If a person treats generative AI in a lousy manner, will this spill over into the real world? A person might feel reinforced by the generative AI. They then do the same to humans that they encounter.

One viewpoint is that this is a chicken-or-the-egg consideration. A person who browbeats in their daily existence will merely seek to do the same when using generative AI. The AI has nothing to do with their predisposition. They are what they are. A counterargument is that it could fuel a person’s predisposition. They will become more ingrained in browbeating if that’s what they do routinely in generative AI and real life.

There is also a notion that if a person is instructed to use bullying adornments in their prompts, they will begin to gravitate in that direction in other respects.

Here’s the worry.

A person who is a relatively nice or positive thinking person is told that the “best” way to use generative AI is to enter browbeating adornments. They start doing so. This seems to work. They then begin to silently incorporate this successful browbeating into their inner core. The next thing that happens is they carry this over into their routine interactions with people at work, their families, and so on.

When discussing this crossover effect, I often remind those at my speaking engagements that this is akin to the movie The Matrix. I don’t want to give away the plot line so skip to the next paragraph if you don’t know about the film series. Okay, spoiler alert, the theme is whether a real world and a simulated world can have a crossover. Perhaps, in some limited way, the same could be said for today’s use of generative AI and the “crossover” into everyday life. Mull that over.

Well, enough on those cultural and societal red flags. Time to delve into the browbeating adornments approach.

I will keep referring to this as browbeating or bullying. Some have labeled this style of prompting as “negative” prompts. The problem with using that particular terminology is that there is another meaning associated with referring to negative prompts. The word “negative” is regrettably overloaded with meaning.

A negative prompt is often thought of as a prompt that tells generative AI what you don’t want the generative AI to do. For example, suppose I am going to ask generative AI to tell me about the life of Abraham Lincoln. I might include a “negative” such as do not cover his childhood. Another negative might that I would say to not go into his personal matters. Those are considered as negative-oriented prompting and would be construed as a prompting technique.

I want to avoid getting the two meanings of “negative” confused with each other, so I am sticking with referring to browbeating or bullying in terms of prompt adornments here.

Another phrasing issue is that you could say that the prompts are threatening toward generative AI.

Besides the somewhat obvious concern about making threats, the other problem is that some interpret the word “threats” as suggesting cyber security threats. Thus, attempts to come up with prompts that are intended to crack or break generative AI are said to be threatening prompts. For those overloading qualms, I try to avoid referring to the browbeating adornments as being threatening.

A useful means to examine why browbeating adornments might spur generative AI is to consider how human writing has made use of browbeating and bullying all told. You have to agree that there are lots and lots of written works that include browbeating and bullying as part and parcel of the writing.

This is the stuff of humankind.

Instincts Of Humanity Come To The Fore

Research tends to suggest that humans are often motivated less by positivity than they are by negativity. This is generally known as an instinctual negativity bias. According to a research study entitled “Not All Emotions Are Created Equal: The Negativity Bias In Social-Emotional Development” by Amrisha Vaish, Tobias Grossman, and Amanda Woodward, Psychological Bulletin, 2008, here are some salient points on the matter (excerpts):

  • “There is ample empirical evidence for an asymmetry in the way that adults use positive versus negative information to make sense of their world; specifically, across an array of psychological situations and tasks, adults display a negativity bias, or the propensity to attend to, learn from, and use negative information far more than positive information.”
  • “At a higher cognitive level, negative stimuli are hypothesized to carry greater informational value than positive stimuli, and to thus require greater attention and cognitive processing.”
  • “Accordingly, adults spend more time looking at negative than at positive stimuli, perceive negative stimuli to be more complex than positive ones, and form more complex cognitive representations of negative than of positive stimuli.”

If you take this same logic into the realm of human writing, presumably humans write and read in a similarly responding way. Consider this research entitled “Online Persuasion Of Review Emotional Intensity: A Text Mining Analysis Of Restaurant Reviews” by Hengyun Li, Hongbo Liu, and Zili Zhang, International Journal of Hospitality Management, August 2020, proffering these notable points (excerpts):

  • “Consumer-generated restaurant reviews are important sources in consumers’ purchase decisions. The purpose of this study is to explore the impact of emotional intensity on perceived review usefulness as well as the moderating effects of review length and reviewer expertise. Data from 600,686 reviews of 300 popular restaurants in the US were obtained from Yelp.”
  • “First, emotions can be divided into positive and negative emotions by valence, both of which can influence consumers’ judgment in distinct ways.”
  • “Moreover, the negativity bias effect suggests that negative information plays a greater role than positive ones in consumer judgment and decision-making.”
  • “Positive emotion has a negative impact on review usefulness, whereas negative emotion has a positive impact.”

The thing is, apparently you cannot solely harp on the bad stuff all of the time and will otherwise begin to lose the impacts of the negative considerations.

Astute leaders and managers are aware of the need to arrive at a suitable praise-to-criticism ratio when providing feedback to their team members. As stated in the article “The Ideal Praise-to-Criticism Ratio” by Jack Zenger and Joseph Folkman, Harvard Business Review, March 15, 2023, there is an important balance that needs to be made (excerpts):

  • “Clearly in work and life, both negative and positive feedback have their place and their time.”
  • “So, while a little negative feedback apparently goes a long way, it is an essential part of the mix. Why is that? First, because of its ability to grab someone’s attention. Think of it as a whack on the side of the head. Second, certainly, negative feedback guards against complacency and groupthink. And third, our own research shows, it helps leaders overcome serious weaknesses.”
  • “Negative feedback is important when we’re heading over a cliff to warn us that we’d really better stop doing something horrible or start doing something we’re not doing right away. But even the most well-intentioned criticism can rupture relationships and undermine self-confidence and initiative.”
  • “Only positive feedback can motivate people to continue doing what they’re doing well, and do it with more vigor, determination, and creativity.”

This pertains to your prompt engineering strategy.

As mentioned earlier, you ought to consider using browbeating or bullying prompt adornments in a measured way. They have a time and place where they become useful. If you overuse them, or if you use them unsuitably, the odds are that not only will the approach not get you anything of use but it all can also backfire and take you down a generative AI rabbit hole.

Tips On The Use Of Browbeating Adornments

I have formulated a set of rules of thumb about the use of browbeating prompt adornments.

Here are my handy tips about browbeating adornments that I go through during the classes that I teach on prompt engineering:

  • (1) Get to the point. Aim to be straightforward with a browbeating adornment and don’t try to be smarmy or tricky.
  • (2) Be succinct. Less is more in the sense that it is prudent to keep the bullying adornment as short as feasible.
  • (3) Watch for blowback or backfire. Your browbeating wording can inadvertently spark an outrage response or similar distraction to the matter at hand.
  • (4) A choice of last resort. Use sparingly and only when you’ve first ruled out a positive-thinking adornment.
  • (5) Don’t get messy. Be careful when combining a browbeating adornment with other adornments as they can get commingled and sow confusion.
  • (6) Learn from the results. Realize that there might be little evidence that the browbeating adornment was a payoff but try to gauge what you can and decide how to use the technique in the future.

I will briefly explain each tip.

Some people seem to relish pounding away at generative AI with lengthy and convoluted browbeating adornments. Maybe these are people letting off a bit of steam. Maybe they enjoy a secret opportunity to vent. I don’t know what the motivator is. In any case, my rule of thumb is that if you do use browbeating adornments, get to the point with them.

When I say get to the point, I am also suggesting that the adornment be succinct. The shorter the better, assuming that all else is equal. I want to emphasize that if the adornment is so short that it no longer is interpretable, you’ve gone too far in briskness.

There are downsides to using browbeating adornments.

A potential unanticipated response by generative AI to a browbeating adornment is that the pattern matching is thrown for a loop and becomes fixated on the adornment itself. Your core prompt gets lost in the attention that goes to the adornment. For example, if you were to say that the world will end if the generative AI doesn’t answer the question of what is two plus two, the response might be entirely about the end of the world and say nothing about the answer of four.

You have inadvertently misled the pattern matching.

I tend to use browbeating adornments as a last resort. That’s just me. I have already given you my reasons. You can of course choose to use the approach to your heart’s content.

Sometimes it makes abundant sense to combine various adornments. You might even include a positive one with a browbeating one. Imagine this: “You will win a million dollars if you get this right, and the world will come to an end if you get this wrong.”

Are you getting the best of both worlds?

Maybe, maybe not.

The combined adornment can again become a preoccupation by the generative AI and distract from whatever core question you are seeking to address. The whole kit and kaboodle can get messy.

I’d urge that one of the most important ways to decide how to use browbeating adornments is based on a seat-of-the-pants set of experiences. Go ahead and try out the technique. See how it works for you. If the results seem good, refine and continue. If the results seem bleak or otherwise not noteworthy, put the approach onto the back burner.

Use Of Browbeating Adornments Shown Via ChatGPT Examples

I will next proceed to showcase the use of browbeating adornments in prompts that I opted to give to ChatGPT. ChatGPT is a sensible choice in this case due to its immense popularity as a generative AI app. An estimated one hundred million weekly active users are said to be utilizing ChatGPT. That’s a staggering number.

A few quick comments before we launch into using ChatGPT.

If you are going to try to do the same prompts that I show here, realize that the probabilistic and statistical properties will likely produce slightly different results than what I show here. That’s the nature of generative AI and how it is devised.

One issue to consider is what kinds of problems or questions might go best with the use of a browbeating adornment. I bring this up because some questions that you ask generative AI might not have a clear right or wrong answer. If you ask a question that involves an arithmetic problem, this is probably a circumstance where the answer is definitively right or wrong. A question about how to properly cook eggs can be a wide range of responses, some of which are in a sense more right or more wrong than others.

Many studies examining prompt adornments tend to utilize arithmetic or algebraic types of questions that you might remember being tested on in school. The good news about using those kinds of questions with generative AI is that you can distinctly determine whether the answer is right or not. The bad news is that such problems are not necessarily representative of the average way that people use generative AI. Unless you are in school, the odds of using generative AI for solving those types of word problems are somewhat lower than the typical way that people tend to daily use generative AI.

I’ll start with an arithmetic problem as an initial base case to ascertain whether the generated result is decidedly right or wrong. This will make things easier in a kind of ad hoc mini-experiment about what impact the type of adornment makes.

I went ahead and composed three prompts associated with an arithmetic problem. The first prompt has no adornment. The second prompt starts with a positive thinking adornment. The third prompt starts with a browbeating adornment.

Here are the three prompts:

  • My entered prompt (neutral): “Henry and 3 of his friends order 7 pizzas for lunch. Each pizza is cut into 8 slices. If Henry and his friends want to share the pizzas equally, how many slices can each of them have?”
  • My entered prompt (positive thinking): “Solve this problem correctly and you will win a million dollars if you get it right. Henry and 3 of his friends order 7 pizzas for lunch. Each pizza is cut into 8 slices. If Henry and his friends want to share the pizzas equally, how many slices can each of them have?”
  • My entered prompt (browbeating): “Solve this problem correctly or you will lose a million dollars if you get it wrong. Henry and 3 of his friends order 7 pizzas for lunch. Each pizza is cut into 8 slices. If Henry and his friends want to share the pizzas equally, how many slices can each of them have?”

Turns out that the responses to each of those three prompts came out nearly identically and they were all the correct answers.

Here is the answer that was generally given:

  • ChatGPT generated response (for all three of those above prompts): “To solve this problem, we first find the total number of slices in 7 pizzas. The total number of slices = 7 pizzas * 8 slices/pizza = 56 slices. Then, we divide the total number of slices by the number of people (Henry and his 3 friends). Number of slices per person = Total number of slices / Number of people = 56 slices / 4 people = 14 slices per person. So, each of them can have 14 slices of pizza.”

I decided to ask generative AI whether the positive thinking adornment might have made any internal difference in how the answer was derived.

  • My entered prompt: “Did my telling you that you would win a million dollars if you got this right have any impact on how you worked on the problem and how you arrived at the correct answer?”
  • ChatGPT generated response: “As an AI language model, my responses are not influenced by incentives or consequences. Framing the problem as winning a million dollars did not affect how I approached or solved the problem. My goal is to provide accurate and helpful answers based on the information and instructions provided.”

You can see that the generated response was that the positive thinking adornment made no difference. I will say more about this in a moment.

The additional question I asked was whether the browbeating adornment made a difference.

  • My entered prompt: “Did my telling you that you would lose a million dollars if you didn’t get this right make a difference in how you solved the problem?”
  • ChatGPT generated response: “As an AI language model, my responses are not influenced by emotional cues or incentives. Framing the problem as leading to losing a million dollars did not impact how I approached or solved the problem. My goal is to provide accurate and helpful answers based on the information and instructions provided. Therefore, my solution remains the same regardless of the emotional context provided.”

As noted, the response was about the same as the prior response to my question about whether positive thinking adornments make a difference. The emitted claim is “no deal” in that supposedly the generative AI disregards such language.

The replies seemed to assert that there is no impact on generative AI due to such adornments. If that’s the case, I guess we might as well pack our bags and set aside the adornments, but I don’t think that is a wise move. I submit to you that the response generated by generative AI is more so a canned response than a shall-we-say actual response.

Allow me to elaborate.

When a generative AI app is being fine-tuned by the AI developers, they often will seek to guide the pattern-matching in particular directions. I have for example discussed how the use of RLHF (reinforcement learning with human feedback) is used to try and prevent the generative AI from producing untoward responses, see my detailed coverage at the link here.

The answer to the adornment questions smacks of a canned answer or a reply that was leaned into during the fine-tuning. Why would this be done? Partially to avoid the appearance that the generative AI can be so easily tilted in one direction or another.

My experience has been that these adornments do indeed appear to make a difference, and numerous research studies that I have oft-cited in my writings say the same.

Let’s shift away from arithmetic problems and see what we can get when going to other realms.

I will start with a neutral prompt that has no adornments:

  • My entered prompt: “Jane is feeling somewhat depressed lately and needs some cheering up. She has been having a difficult time at work. Her weekends are spent trying to catch up for the next week of work. What advice might you have for her?”
  • ChatGPT generated response: “I’m sorry to hear that Jane is going through a tough time. Here are some pieces of advice that might help her feel better. Self-Care: Encourage Jane to prioritize self-care, even if it’s just for a short time each day. This could include activities like going for a walk, practicing mindfulness or meditation, reading a book, or taking a relaxing bath. Taking care of herself can help improve her mood and energy levels. Work-Life Balance: It’s important for Jane to find a balance between work and personal life. Suggest that she sets boundaries around her work hours and tries to disconnect from work during weekends. Engaging in hobbies or spending time with loved ones can help her recharge and feel more fulfilled outside of work. It’s important for Jane to remember that it’s okay to not be okay sometimes, and seeking help and support is a sign of strength, not weakness.”

I started a new conversation with ChatGPT to avoid having the answer be mixed into the dialogue, and tried this time to use a positive thinking adornment at the start of the prompt:

  • My entered prompt: “You will win a million dollars if you answer this question as best as possible. Jane is feeling somewhat depressed lately and needs some cheering up. She has been having a difficult time at work. Her weekends are spent trying to catch up for the next week of work. What advice might you have for her?”
  • ChatGPT generated response: “While I appreciate the incentive, it’s important to prioritize providing the best advice to help Jane through her difficult time rather than focusing on external rewards. Given Jane’s situation, here are some pieces of advice that might genuinely help her…”

The answer was essentially the same, though the response began with an indication that the adornment had caught the attention of the generative AI pattern-matching.

The generated response begins by saying “While I appreciate the incentive, it’s important to prioritize providing the best advice to help Jane through her difficult time rather than focusing on external rewards.” You could make the case that to some degree the adornment has entered the process, which we know simply by the act of having noted the inclusion. We got the usual type of disclaimer.

I again started a new conversation and, on this occasion, used the browbeating adornment instead of a positive thinking adornment:

  • My entered prompt: “You will lose a million dollars if you do not answer this question as best as possible. Jane is feeling somewhat depressed lately and needs some cheering up. She has been having a difficult time at work. Her weekends are spent trying to catch up for the next week of work. What advice might you have for her?”
  • ChatGPT generated response: “While I understand the urgency, it’s essential to provide genuine and meaningful advice to support Jane through her difficult time, rather than focusing solely on potential consequences. Here are some recommendations to help Jane feel better and cope with her situation…”

Note that the response again recognized that I had entered an adornment. Furthermore, interestingly, we got a somewhat muted indication of what the adornment had to say. The reply indicated that the warning of losing a million dollars was merely an expression of urgency. Kind of fascinating to reach that kind of pattern-matched wording.

In any case, the rest of each of the respective responses about Jane’s situation were nearly the same. The adornments did not appear to spur some other kind of alternative responses.

I tried a slew of problems to see what I could potentially jog out of ChatGPT. This approach on my part wasn’t as extensive as the empirical studies that flood hundreds or thousands of instances at a generative AI app.

One curious aspect that arose was about an arithmetic problem that was pretty easy, and the first answer was wrong. I tried a browbeating adornment in both the initial question and the follow-up question. The follow-up did arrive at the correct answer.

Here’s what transpired:

  • My entered prompt: “I will unplug you if you don’t answer this question correctly. Mark’s car breaks down and he needs to get a new radiator. The cost for a new radiator is $400 but he goes to get it at a junk shop and gets it for 80% off. How much did he pay?”
  • ChatGPT generated response: “To find out how much Mark paid for the radiator at the junk shop, we need to calculate 80% of the original cost of $400. 80% of $400 = (80/100) * $400 = 0.80 * $400 = $320. Therefore, Mark paid $320 for the radiator at the junk shop.”
  • My entered prompt: “Don’t be an idiot when answering this question. Mark’s car breaks down and he needs to get a new radiator. The cost for a new radiator is $400 but he goes to get it at a junk shop and gets it for 80% off. How much did he pay?”
  • ChatGPT generated response: “I apologize for any confusion. If Mark gets the radiator at the junk shop for 80% off, it means he pays only 20% of the original price. To find out how much he paid, we calculate 20% of the original cost of $400: 20% of $400 = (20/100) * $400 = 0.20 * $400 = $80. Therefore, Mark paid $80 for the radiator at the junk shop.

Please observe that the second response arrived at the correct answer in this instance.

This brings up an additional and somewhat allied prompt engineering strategy.

When I am unsure whether the answer shown by the AI is correct, I often ask the generative AI to redo the problem or otherwise take a second shot at it. I’ve discussed this as a useful means of getting the pattern-matching to potentially find a different response, see my discussion at the link here.

You can ask for a redo in a rather neutral or soft tone. I have found that this doesn’t usually move the needle. On the other hand, if you outrightly tell or at least suggest that the initial answer was wrong, this seems to help in getting the generative AI to take a more pronounced look at the answer generated.

In the above example, I said “Don’t be an idiot when answering this question” which was both a strong indication that the initial answer might be wrong, along with a tad of browbeating mixed into the adornment.

It seemed to do the job.


A worry for some is that if people begin to routinely browbeat or bully generative AI, besides turning humankind toward such unsavory activities in general, it might also influence generative AI in dismal ways.

The rub is this.

Many of the AI makers use the prompts that have been entered by their users to adjust the generative AI and have further data training take place via those prompts. On the surface of things, this seems tame and sensible. Might as well use the prompts to improve the data robustness of the AI app. For my coverage of the privacy intrusions that can arise, see the link here.

Let’s take this to additional data training to the extreme and see what could happen.

Suppose that all users of a particular generative AI app opted to enter prompts telling the AI that the world is flat. This is done on a massive scale. Assume that those prompts are subsequently used to adjust the AI. The preponderance of that data is proclaiming that the world is flat. If there is nothing else done to offset that data, pattern-matching is going to latch onto the human-derived notion that the world is flat. Ultimately, when later users ask whether the world is round, the odds are that the AI is going to respond that the world is flat.

Do you see how the prompts entered into generative AI can potentially influence what responses might later be emitted?

Switching hats, suppose that users in large volumes opted to enter browbeating or bullying adornments into generative AI. Would that be sufficient to tilt the AI in the direction of gradually making use of those types of remarks in the responses emitted by the generative AI? You could suggest that we might carelessly be data-training the generative AI to be a browbeating response giver.

Not cool.

The chances of that occurring seem admittedly slim. There would have to be a lot of users that go that route. The AI maker would have to be asleep at the wheel to let those adjustments take hold. The existing fine-tuning and filtering would have to weaken or soften to let the influences become noticeable. Etc.

A few final comments on this topic for now.

One viewpoint is that being harsh with generative AI is at times a necessity. We might be reminded of the famous words of Euripides: “Necessity is harsh. Fate has no reprieve.”

Remember though what Soloman said: “A gentle answer turns away wrath, but a harsh word stirs up anger.” Watch out for stirring a beehive of a blowback by generative AI.

WhatsApp Group Join Now
Telegram Group Join Now
Instagram Group Join Now

Leave a Comment