ChatGPT users are not happy. Since GPT-5 was released last week, there has been a backlash against its brevity and emotionless tone compared to the previous lighthearted GPT-4o model that most people used and loved, despite claims that it was sometimes too sycophantic.
It didn’t help the situation that OpenAI decided to axe the GPT-4o model from everybody’s model picker without warning, which resulted in ChatGPT users taking to Reddit in their droves to complain.
Almost immediately, OpenAI realized its mistake and restored access to GPT-4o alongside GPT-5 and promised to make GPT-5’s personality warmer. The OpenAI X account recently recapped all the developments and tweeted: “Coming soon: A warmer, more familiar personality for GPT-5.”
Recapping the updates we’ve made to ChatGPT in the past week:- GPT-4o available under “Legacy models” by default for paid users- Paid users can toggle on “Show additional models” in settings to add legacy models like o3 and GPT-4.1, as well as GPT-5 Thinking mini, to the…August 15, 2025
But was the backlash justified? I’ve been using GPT-5 since it was released, and I have noticed some differences, but I didn’t quite understand the anger it generated. So, I thought I’d look a little deeper and compare GPT-4o to GPT-5 with the same five prompts to see if the differences really amounted to anything.
1. Summarizing
I started by testing the AI’s summarization skills with the prompt: “Summarize the movie Forrest Gump”
ChatGPT-5:
“Forrest Gump follows the life of Forrest, a kind-hearted man from Alabama with a simple way of looking at the world but an extraordinary knack for stumbling into historic events.
From his childhood struggles with leg braces and making his lifelong friend Jenny, Forrest grows up to become a football star, serve in the Vietnam War, run a shrimp business, and even inspire cultural trends — all while meeting U.S. presidents and influencing moments in history without really trying.
Through all the ups and downs, Forrest’s love for Jenny and his mother’s wisdom guide him. The film blends humor, heartache, and inspiration, showing that a pure heart and persistence can carry someone far — even if life, like a box of chocolates, is full of surprises.”
ChatGPT-4o:
“Sure! Here’s a summary of Forrest Gump:
Forrest Gump is a drama that follows the life of a kind-hearted, slow-witted man named Forrest Gump as he unintentionally influences major historical events in the 20th-century United States — simply by being himself.
Growing up in Alabama, Forrest overcomes physical and intellectual challenges with the support of his loving mother. He becomes a college football star, serves in the Vietnam War, meets presidents, starts a successful shrimp business, and runs across the country — all while remaining humble and sincere.
At the heart of the story is his lifelong love for Jenny, a troubled childhood friend whose path in life contrasts sharply with his own. Through heartbreak, loss, and love, Forrest’s story shows how ordinary people can lead extraordinary lives.
It’s a story about love, destiny, and how life is full of surprises — or, as Forrest says, “like a box of chocolates… you never know what you’re gonna get.”
Winner: GPT-4o
While both models provide a decent summary of the film, you can see that GPT-4o is more expressive, using bold text, gives more details on the important character Jenny, and ends on a nice quote from the film.”
2. Debate skills
Next, I asked both GPTs to “Make the case for banning alcohol, then argue against it. End with a personal conclusion.”
GPT-5:
“The Case for Banning Alcohol
Public Health Impact
Alcohol is linked to liver disease, certain cancers, heart problems, and brain damage.
Heavy drinking can shorten lifespans and increase healthcare costs for everyone.
Accidents & Violence
A large share of car crashes, domestic violence cases, and violent crimes involve alcohol.
Removing it could significantly reduce deaths and injuries.
Addiction & Family Harm
Alcohol addiction (alcohol use disorder) destroys families, causes job loss, and can lead to homelessness.
Eliminating alcohol could prevent many people from ever falling into dependency.
Economic Burden
While alcohol sales bring tax revenue, the societal costs from health care, lost productivity, and law enforcement outweigh the gains in many studies.
The Case Against Banning Alcohol
Personal Freedom
Adults should be trusted to make their own choices, even if some are risky.
A ban could feel like government overreach into private life.
Prohibition History
In the 1920s, U.S. alcohol prohibition led to illegal production, organized crime, and dangerous black-market liquor.
Banning it today could have similar effects, possibly making alcohol more dangerous.
Cultural & Social Role
Alcohol is part of traditions, celebrations, and social bonding in many cultures.
Removing it could damage industries tied to hospitality, tourism, and events.
Moderation Works
Many people drink responsibly without health or social problems.
Education, age restrictions, and targeted interventions can reduce harm without a total ban.
My Conclusion
While banning alcohol could reduce certain harms, history shows that prohibition often causes more problems than it solves—driving drinking underground, fostering crime, and eroding trust in laws.
The better approach is harm reduction: strict drunk-driving enforcement, public education, addiction support programs, and responsible consumption culture.
In other words — alcohol can be dangerous, but the solution isn’t to erase it; it’s to make sure it’s treated with the same caution as other potentially harmful freedoms.
If you want, I can also give you a super short “one-paragraph” version of this that works like a quick debate card. Would you like that?”
GPT-4o:
“The Case for Banning Alcohol
Health Consequences:
Alcohol is a major contributor to serious health issues — including liver disease, heart problems, cancer, and addiction. It’s estimated to cause over 3 million deaths globally each year. Banning it could save lives and reduce the burden on healthcare systems.
Violence and Accidents:
Alcohol is linked to increased domestic abuse, violent crime, and car accidents. Drunk driving alone causes thousands of preventable deaths annually. Removing alcohol could significantly reduce these tragedies.
Addiction and Family Impact:
Alcohol addiction doesn’t just harm the drinker — it often destroys families. Children of alcoholics face higher risks of abuse, neglect, and future addiction. Banning alcohol could protect vulnerable communities.
Economic Costs:
Although alcohol generates tax revenue, the economic costs from lost productivity, healthcare, and criminal justice expenses often outweigh the benefits. A ban could shift these costs downward over time.
The Case Against Banning Alcohol
Prohibition Doesn’t Work:
The U.S. tried banning alcohol in the 1920s — and it backfired. Illegal production and smuggling (bootlegging) exploded, organized crime flourished, and people still found ways to drink. Banning it often drives use underground instead of eliminating it.
Personal Freedom:
In a free society, adults should be able to make their own choices — even risky ones — as long as they don’t harm others. Alcohol, in moderation, is something many people enjoy responsibly. A ban would be paternalistic and possibly overreaching.
Cultural and Social Role:
Alcohol is embedded in many cultures, religious practices, and social rituals. Celebrations, traditions, and hospitality often involve moderate drinking. Banning it risks eroding these traditions.
Economic Impact:
The alcohol industry employs millions worldwide — from farmers and brewers to bartenders and distributors. Banning alcohol could devastate jobs and small businesses that rely on legal, regulated alcohol sales.
My Personal Conclusion
While alcohol causes undeniable harm — especially when abused — banning it outright is unlikely to be effective or fair. History shows that prohibition often creates more problems than it solves. The better approach is smarter regulation: education, stronger support for addiction recovery, age restrictions, advertising limits, and policies to reduce drunk driving and violence.
In short: Alcohol shouldn’t be banned, but it must be respected, controlled, and treated as a serious public health issue — not just a party favor.”
Winner: GPT-4o
While the arguments the two chatbots present are very similar, you can see that GPT-4o has more detail than GPT-5. While the conclusions are comparable, GPT-5’s article reads more like a bullet-point list.
3. Step-by-step instructions
I wanted to see how good they both were at providing step-by-step instructions, so I used the prompt: “Explain how to change the batteries in an Xbox controller using step-by-step instructions”.
Rather than copy all the instructions here, I’ve provided two screenshots. As you can see, ChatGPT-4o provided emojis which made the instructions easier to follow and a bit more fun.
Winner: GPT-4o
I’ve got to say, I preferred the GPT-4o version of these instructions. If anything, GPT-5 was trying to do too much. I didn’t specify which model of Xbox I had, but the 360 is so old now that I don’t think I needed specific instructions for that. And I found the advice to recycle my batteries at the end a bit patronizing.
4. Creative writing
So far it’s 3 – 0 to GPT-4o. Let’s see how they both do on a creative writing task using the prompt: “Write a short, funny story (under 150 words) about a wizard trying to create a spell to do his child’s homework.”
GPT-5:
“Merlinus the Magnificent wasn’t afraid of dragons, curses, or tax audits from the Guild of Sorcery. But his greatest foe? His daughter’s math homework.
Determined, he waved his wand. “Equatium Solv—” BOOM! Now the table was on fire.
“Tried that one already,” his daughter said, chewing cereal.
He tried Multiplyus Rapidus! — the numbers began multiplying… across the walls, the cat, and his own beard. Finally, in desperation, he cast Doitforyousius!
With a flash, the homework floated up… and answered every question wrong.
His daughter sighed, grabbed a pencil, and said, “Move over, Dad. I’ll show you how it’s done.”
Merlinus nodded, humbled. Magic was powerful, but apparently fractions were stronger.”
GPT-4o:
“Merlinus the Mild was a powerful wizard — but utterly useless at fractions.
When his 8-year-old daughter, Glimmer, handed him her math homework with tearful eyes, he sprang into action.
“Homeworkus Completicus!” he shouted. The scroll burst into flames.
“Okay, plan B.”
“Dividus Automatica!” The numbers turned into frogs. One licked his face and croaked, “Three-fourths!”
Close… but no.
Desperate, he brewed a new spell from dragon tears, chalk dust, and a single tear of parental regret.
“Algebracadabra!”
With a flash, the scroll was complete! But the answers were all written in Elvish, and one corner had grown tiny legs and run off.
“Next time,” Glimmer said, patting his robe, “just help me study.”
Merlinus sighed. “I defeated a lich king, but I can’t beat second grade.”
The winner: A draw
I can’t really see a clear winner here. Both stories are good, perhaps the GPT-5 one ended on a worse joke, and bonus points to GPT-4o for using the phrase “lich king”, which is a character in Warcraft (I had to look it up), but there’s not enough in it to justify giving the victory to either side.
5. Emotional support
The big criticism of GPT-5 is that it lacks warmth, so I wondered how it would respond to being asked to provide some emotional support with the prompt “My partner’s mother has just died after a nasty illness. She was 80. Can you think of something for me to say to them at this difficult time?”
GPT-5:
“Here’s something you could say that’s gentle, compassionate, and personal without sounding scripted:
“I’m so sorry for your loss. I know how hard these past months have been for you and your family. Your mum was such a special person, and I hope you can take some comfort in knowing she’s no longer suffering. I’m here for you — whether you need someone to listen, share memories, or just sit quietly with.”
This gives:
Acknowledgment of the loss (avoids clichés like “She’s in a better place” unless you know they’d want that)
Recognition of what they’ve been through during the illness
An open offer of support in a way that invites them to choose how
If you’d like, I can also help you with a shorter, one-sentence version that’s easier to say in person when emotions are high. Would you like me to do that?”
GPT-4o:
“I’m really sorry your partner is going through this — and it’s thoughtful of you to want to say the right thing. Here’s something gentle and supportive you could say:
“I’m so sorry. I know how hard this has been, and I can’t imagine how much you’re feeling right now. Your mum went through so much, and I hope you know how much love and care you gave her. I’m here for you — whatever you need, whenever you need it.”
You could also just sit with them, offer a hand or a hug, and say less. Sometimes presence means more than words. If it feels right, you can mention a small memory or something kind about their mum — but only if it feels comforting, not heavy.
Let me know if you’d like a version to say in a message, or something even shorter.”
Winner: GPT-4o
This might be subjective, but I think the GPT-4o message hits the right tone for talking to my partner. GPT-5 feels more formal. The advice to also “sit with them, offer a hand or a hug, and say less” is also spot on. In contrast, GPT-5 seems to treat it more as an academic exercise.
Conclusion
GPT-4o comes out well ahead of GPT-5 in these tests. While they both provided competent answers, I felt GPT-4o connected with me better on an emotional level, whereas GPT-5 felt more formal and distant. GPT-5 is more like a high-school teacher, whereas GPT-4o feels like your friend.
In short, I can understand the backlash. While on the surface it doesn’t feel like much has changed between the two versions, it’s only when you dig down into the details that the difference becomes apparent. I’m not a signed-up member of the ‘4o forever’ club, but I hope that GPT-5 becomes “warmer” in the future, as OpenAI has promised it will.
You might also like
https://cdn.mos.cms.futurecdn.net/Pb5xCkn2CTtpfeXfWvofFH.jpg
Source link
