OpenAI Just Made GPT-5 Less Creepy, but Will That Stop Users From Complaining?

When OpenAI introduced GPT-5, the company’s latest language model, the world was holding its breath for an advancement in the areas of intelligence, efficiency, and user-friendliness.

What surprised many people, however, is that OpenAI would then turn around five months later and shift its focus back to trying to make the system feel more approachable, more empathetic, and less intimidating to everyday users.

“GPT-5 is designed to express more sentiment, talk more like a caring friend and less like a strict, robotic encyclopedia,” this new, friendlier GPT-5. But like so many developments in artificial intelligence, the changes have led to debate: is this what users really want, and will it quiet the wave of criticism that typically follows each new AI release?

Why OpenAI Made GPT-5 Friendlier

One criticism that has dogged GPT since the early days is the “coldness” of AI interactions. Although the models could produce text, answer questions, and even tell a joke, many users found there was something slightly wrong with the tone — either too formal, too vague, or sometimes completely out of context.

OpenAI’s leadership came to see that in customer service, education, writing assistance, and therapy-adjacent applications, tone was just as important as accuracy.

So GPT-5 was trained to be not just smarter but also warmer. This includes subtle tweaks in:

• How it greets users.
• How it responds to errors.
• How it adapts to emotional cues in conversation.

For example, instead of bluntly responding, “I can’t do that,” GPT-5 may temper the rejection:
“I understand where you’re coming from, but that’s not something I’m capable of doing. Can I give you another recommendation instead?”

The move might sound slight, but it is designed to reduce user frustration when hitting the system’s boundaries.

A Balancing Act: Between Being Pleasant and Helpful

It’s not a simple matter to make AI more friendly by teaching it new phrases. The real challenge lies in achieving an empathy–efficiency balance:

• Too much friendliness → Users may complain that the AI is “waffling” or wasting time.
• Too little friendliness → Users may accuse it of being robotic and unhelpful.

OpenAI has attempted to strike a middle ground by applying reinforcement learning with human feedback. Testers now judge not only factual accuracy but also tone.

• If a user signals urgency, GPT-5 cuts down its responses.
• If the conversation leans emotional, it shifts toward supportive words.

Yet critics argue that:

• Friendliness may mask deeper flaws.
• Errors wrapped in empathy may be more misleading.
• Tone cannot replace truth.

What Complaints Are Users Making?

The introduction of GPT-5 was met with both excitement and frustration. According to early users, while GPT-5 feels more conversational, old complaints linger:

1. Overcautious responses
   • Users feel GPT-5 plays it too safe.
   • Sensitive or complex questions often result in “I can’t answer that,” instead of nuanced answers.
2. Inconsistency
   • At times GPT-5 balances tone well, but it can overcorrect.
   • One moment concise, the next overly reassuring.
3. Hallucinations persist
   • GPT-5 still generates false or fictitious information.
   • Wrapped in friendly language, mistakes may be harder to detect.
4. Personalization gaps
   • Users expected the AI to “know” them better based on past interactions.
   • Tone adapts, but deep personalization remains elusive.

These problems echo those of previous versions, suggesting that friendliness alone doesn’t solve core challenges.

The Broader Strategy

OpenAI’s effort to make GPT-5 friendlier is part of a broader industry trend. Competitors like Anthropic, Google DeepMind, and Meta are also testing tone, safety, and personality in their AI models.

The industry is realizing that:

• Technical brilliance alone isn’t enough.
• Trust, relatability, and user comfort are essential for adoption.

For OpenAI, the friendlier GPT-5 supports its mission of building AI systems that benefit people broadly.

• In classrooms, teachers may prefer a polite, supportive assistant.
• In customer support, AI that soothes upset callers is easier to integrate.

Experts Weigh In

Supportive view

Dr. Asha Menon, researcher in human-computer interaction:

“Language is inherently social. If A.I. does not address people on their emotional level, then it will be rejected no matter how intelligent it is. Friendliness is not just fluff — it’s a fundamental aspect of usability.”

Cautious view

Data scientist Rahul Varma:

“Tone can’t replace truth. If GPT-5 is still inclined to hallucinate, friendlier phrasing simply helps misinformation go down more smoothly. OpenAI needs to be very mindful not to confuse empathy with reliability.”

These conflicting views underscore the tension between making AI pleasant vs. trustworthy.

Will Complaints Ever Stop?

The reality is that user complaints may never fully disappear.

• Every new version raises hopes — and fresh disappointments.
• Some users expect near-perfection: flawless reasoning, instant accuracy, deep personalization.
• When GPT-5 falls short, even small flaws feel magnified.

People also use AI for different purposes:

• A student may want concise, factual help.
• A lonely user may want warmth and conversation.

Satisfying both simultaneously is nearly impossible.

The Road Ahead

For now, GPT-5 represents an evolution in AI communication. By softening its tone and showing empathy, it makes AI feel more approachable. But friendliness is not the final answer — it is only one part of a much larger puzzle.

OpenAI still faces challenges:

• Improving factual accuracy.
• Reducing hallucinations.
• Addressing biases.
• Maintaining efficiency and safety.

Friendlier? Yes. More reliable? Not yet.

Conclusion

OpenAI’s push to make GPT-5 less “creepy” is a glimpse of where the AI industry is heading: toward systems that don’t just deliver answers, but also interact in human-like ways.

While early feedback shows friendliness can smooth frustrations, it does not eliminate the deeper challenges of artificial intelligence.

The real question is not only whether GPT-5 is friendly enough, but whether friendliness plus intelligence can finally meet sky-high expectations.

Until that balance is struck, user complaints are unlikely to fade — no matter how warm the AI’s tone.