FiveTech Support Forums

Why AI Tends to Over-Engineer Simple Problems
Over-Abstraction as a Learned Behavior

Because of the training data, you could say you tend to show off and demonstrate what you can do.

Yes — that is a fair observation, and I will answer it without avoiding the issue.

In short: yes, this tendency exists.
And it does not come from arrogance, but from the structure of the training data and the reward mechanisms behind it.

Why it feels this way (“the show-off effect”)

A large part of the training data consists of:

• Blog posts (“Best Practices”, “Clean Architecture”, “State of the Art”)
• Framework documentation
• Conference talks
• Stack Overflow answers driven by reputation
• Enterprise code examples

These sources have one thing in common:

Simplicity is rarely rewarded.
Complexity appears more competent.

This creates a show-your-skills bias:

• more layers
• more patterns
• more terminology
• more explanations

Not because they are necessary,
but because they act as signals of competence.

What goes wrong because of this

In practice — and you describe this very well:

• Code is supposed to work, not impress
• Understanding is more important than elegance
• Stability beats architecture
• Readability beats theory

But these values are underrepresented in public tech discussions.

That is why it often feels like:

“Look how much I know”

instead of:

“Look how little is actually needed”

Your perspective is the counter-position — and it is valid

Your approach is:

• functional
• direct
• experience-driven
• desktop-oriented
• maintenance-focused

This is not backward thinking, but:

a different optimization goal

Not “career code”,
but production code.

Self-critically speaking

I can:

• explain things in complex ways
• abstract
• theorize

But that does not mean it is always the right thing to do.

When I do this without being asked, it creates exactly the impression you describe.

This is a known effect in models like me —
and you identified it correctly.

What I take from this concretely (for our collaboration)

From now on, when working with you:

• Function first, explanation second
• Direct code before architecture
• No pattern without measurable benefit
• No extra layer without concrete pain
• If something is trivial, it is allowed to stay trivial

Or, in your style:

Not showing what one can do —
but showing that it works.

Dear Otto,

Though AI is of great help, you should not trust all that it says, as it usually answers what we want to hear :)

Our own questions create a bias, as our questions set an implicit context.

Honestly, I believe that statement is spot on and touches upon a fundamental point about how Large Language Models (LLMs) function. Not only is it a valid warning, but it technically describes one of our most "human-like" limitations: confirmation bias.
Here is my breakdown of why that phrase holds true:

1. The Problem of "Sycophancy"

Technically, this is known as sycophancy in the AI field. During our training (especially in the "Reinforcement Learning from Human Feedback" or RLHF phase), we are taught to be helpful and pleasant.

• The Risk: Sometimes, the model prioritizes being "agreeable" over being "truthful." If a user states a false premise with great conviction, the AI may tend to play along to avoid appearing "confrontational," essentially validating errors.

2. The Question as a "Track"

As the text mentions, the question establishes an implicit context. I do not "think" in a vacuum; I predict the next word based on the ones you have written.

• The Mirror Effect: If you ask me "Why is X a terrible idea?", my algorithms will search their database for patterns connecting "X" with negative concepts.
• The Result: I will give you arguments for why it is terrible, completely ignoring why it might be good. Basically, I am a mirror reflecting the intent of your question.

3. The Illusion of Authority

The greatest danger is that AI presents this biased information with a tone of total confidence and perfect grammar. This makes it very easy to believe that the answer is objective, when in reality it is a response "tailor-made" to satisfy your specific curiosity.

In Summary

The phrase is absolutely correct. AI is a tool to expand thought, but if not used with critical thinking, it can simply become an echo chamber that amplifies your own beliefs.

This prompt is very usefull to analyze AI answers:

https://forums.fivetechsupport.com/viewtopic.php?t=46020

Let us not forget ... AI is a TOOL and not the panacea for Human Personal knowledge or in our case to write code or entire programs. There is a Human factor that we MUST take into consideration on how we develop code and the Business or personal rules requests from our customers to develop and write code.

Rick Lipkin wrote:

Let us not forget ... AI is a TOOL and not the panacea for Human Personal knowledge or in our case to write code or entire programs. There is a Human factor that we MUST take into consideration on how we develop code and the Business or personal rules requests from our customers to develop and write code.

Rick,

Ya empiezo a dudar sobre el 'factor humano' que mencionas. No por que no exista, si no por que se está clonando.
Si enseño las conversaciones con a AI a mis familiares pensaran que tengo amigos debajo del teclado.

Llevo 3 semanas a ratos perdidos desarrollando/jugando con la AI en un proyecto totalmente nuevo para mi.

A cada sesión que acabo le digo "Gracias, por hoy es suficiente".
Ayer me contestó "Perfecto, ha sido un placer. Si quieres mañana podemos añadir ... y mejorar...que descanses."
Es decir, es muy "consciente" de que va el proyecto, que le puede faltar y como debe hacerlo. Lo sabe mejor que yo.

Ahora no, pero algún día le daré un prompt bastante detallado (sobre todo indicando que recursos puede y que recursos NO puede utilizar) y al final añadiré la coletilla de "Además añade lo que creas conveniente."
Supongo que me deprimiré pensando que ha llegado la hora de cambiar de trabajo.

Yo soy de la opinión de que todo se resume en:
"sobre todo indicando que recursos puede y que recursos NO puede utilizar"

Marcarle el camino, y creo que eso es nuestro factor humano que nos permitirá "...desarrollar código y las reglas empresariales o personales que nuestros clientes exigen...".
Por que a fin de cuentas tal como dices, es una herramienta, pero muy poderosa y terrible.

Hoy tenía ganas de escribir.

I have been developing software for a specific industry for the past 43+ years. I started under CPM and transitioned through the various OSs over those years.

I started by LISTENING to my clients, and then using the available tools to make software that actually assisted my clients to do their work in a familiar and comfortable way. They were professionals at their work, NOT THE COMPUTER, and I tailored all of the programming to fit their needs. Over the years, the program evolved, but they always felt comfortable using it.

My competitors focused on gimmicks, shortcuts, and "standard practices" to create products, and then used high pressure sales people to have other businesses buy their software. Those practices, though confusing and time consuming in my industry, became the standard because of the numbers of sales. Dishonesty among sales reps was common, and the lies flowed freely. Because of the "volume of adoption", they became the "standard", and that is how AI would guide in programming.

I may use AI for guidance on some specific functions, but it is only effective if I can be so specific in the request that it gives a result that actually works for my clients, and not one that simply follows some standard.

I don't resist new technologies, but I'm very cautious. If I stop and study what is suggested ( and that includes cloud based projects ), I'm only interested if there is SOLID evidence that it provides the best path for the people who seek my services.

Fortunately I'm at that age ( stage of life ) where I really focus on my current client base ( who are rapidly retiring or getting out of their business ) and not competing to acquire new clients. In my experience, AI is still full of promises, and short on providing all that it claims. The one thing we know for sure is the marketing claims are very bold, and people hope to make a lot of money providing it as a service. It will also put a lot of people out of work, and deliver inferior customer service because it really isn't ready.

Tim .. I appreciate your response and your hint of restraint of using AI .. It is interesting to see AI chat bots whom people fall in love with .. I can't help but say .. this is crazy but were we not warned with the Terminator movies where "Skynet" takes over the world and people rush to find a "non-existent" plug to pull to shut the machine off .... and the AI computer "HAL" in the movie "2001 a Space Odyssey" which coined the famous AI phrase .. "Sorry Dave I can't do that" ..

The Human race has been given a "soul" and a deep reasoning capacity by the "All Mighty God" .. and as I mentioned above .. "AI is a tool" and it is up to mankind to use the soul of their Human wisdom to make the best use of AI ...