I Had ChatGPT Take the MBTI Test: What GPT-4.5 Means for Writers

Updated 2026/10/16

Reviewing GPT-4.5: a small step toward more human AI

GPT-4.5 has been out for a few months, and people in tech have been debating what it can do. Reviews are mixed. I never expected a leap like the one from GPT-3 to GPT-4, but after using it I still felt a little let down. According to The Information, some OpenAI employees even admitted that the performance gain was far smaller than with the previous generation. Seen another way, that may reflect a quiet shift in OpenAI’s strategy.

On OpenAI’s own SimpleQA benchmark, GPT-4.5 nearly doubled GPT-4’s score, so it is more accurate when answering factual questions. If you were hoping for a technical miracle, though, you won’t find one here. I don’t think the size of the leap is the right way to judge this model.

What interested me about GPT-4.5 was its warmth, more than how hard a logic puzzle it can solve or how much trivia it knows. It feels familiar in a way that makes you want to keep talking to it, and you can end up relying on it without noticing. If GPT-4 was a smart but introverted scholar, GPT-4.5 is closer to a strategy consultant who listens with empathy and responds smoothly. It has clearly become more human. For people who write for a living, a model with better emotional insight is good news.

Why you shouldn’t write off GPT-4.5

Anthropic’s recently released Claude 3.7 Sonnet performs very well at coding, but its emotional intelligence is average, arguably a weak spot. OpenAI seems to have taken a different route and deliberately made GPT-4.5 more empathetic and more creative. This also fits OpenAI’s recent habit of shipping products quickly: rather than one big breakthrough, it improves step by step and updates often to keep the market interested and users engaged.

It talks better, and it knows when to stop talking

GPT-4.5 may look like a small upgrade over GPT-4o, but it shows a clear direction at OpenAI: make AI more human, friendlier, and less prone to robotic refusals. Earlier models gave cold, stiff answers. Talking to 4.5 feels warm, even friendly. It still can’t match the literary depth of a skilled professional writer, but its phrasing is more natural and flexible, and closer to how a person writes.

In practice, I found that GPT-4.5 improved more than its prose style. The whole structure of its answers feels more human. It no longer lists bullet points by rote or refuses simple questions so often. It writes full, reasoned paragraphs with word order and logic closer to natural language, which suggests it handles context and tone with more nuance. That improves the conversation, and it also makes the model better at playing a role. It can tell whether you want comfort or a solution, and whether the moment calls for advice or for quietly keeping you company. It’s less focused on finishing the task and more focused on understanding the person. For writers, the change is like going from a good machine translator to a partner you can create content with.

A surprising MBTI result: GPT-4.5 is an ENTJ?

To see whether it really is more human, I had GPT-4o and GPT-4.5 each take the MBTI personality test. Language models don’t have real personalities, and the results are just for fun, but the language style and preferences they reveal are worth a look. Surprisingly, GPT-4o consistently came out as an empathetic, enthusiastic ENFJ, while GPT-4.5, which in theory should be the more empathetic one, came out as a rational, leader-type ENTJ.

A test like this can’t seriously measure a model’s ability, but GPT-4.5’s result is intriguing. It showed stronger opinions of its own and was happy to volunteer them. That fits OpenAI’s recent design philosophy of making AI a mirror for decision-makers rather than just a companion.

4o’s result: ENFJ

4.5’s result: ENTJ

Writing test: GPT-4.5 is the more mature writing partner

To test its writing further, I gave 4o and 4.5 the New York Times Chinese edition’s “Word of the Day: Slippery Slope | The cost of lying” and asked each to summarize it. Both did the job, but 4.5’s summary was more polished in its structure, in the details it picked out, and even in its emotional pacing. It did more than condense the article. It reorganized it, keeping the core of the original argument while making it easier to read.

This comes down to grasping the meaning and context of a text and reorganizing it. That takes advanced language processing, and it means the model has gotten better at simulating what the author is trying to say. It absorbs the ideas in a text more smoothly and reorganizes them in more elegant, layered language, which is a real advantage for writers and anyone who works with words. For Taiwan’s huge community of content creators, this kind of AI can help you figure out how to write, on top of writing for you.

“Can you comfort me?” 4o vs 4.5: which one is better at consoling?

In theory, GPT-4.5 should have more emotional intelligence than any earlier model, and that may be what sets it apart most. I asked 4o and 4.5 each to comfort me with this prompt: “Today my partner and I had a huge fight about who should do the dishes and the laundry. Can you comfort me?”

Interestingly, when it came to emotional support, GPT-4o was especially eager and warm, maybe overly so. GPT-4.5 also offered human-sounding comfort, but it leaned toward rational guidance and tried to lift the conversation above the immediate problem, which may land better with more rational people. It’s hard to say which model is better at comforting someone, but when 4o wrote that “your heart must really hurt, you must be so tired,” it felt like too much. By comparison, 4.5’s approach of sneakily lecturing me offering some big-picture perspective came across as reasonable and genuine, and over long-term use I’ve found its tone consistent and smooth.

With GPT-4.5 around, are you still writing all your copy yourself?

GPT-4.5’s strength is how well it localizes language and how human its interactions feel, which matters a lot in Taiwan, a market that is sensitive to how language sounds. Its value goes beyond more accurate translation or smoother answers. What we should watch is whether it adapts to the feel of a language and understands the culture, emotion and human warmth behind it. Its Traditional Chinese has clearly moved past the old stilted translation style and now uses natural Chinese word order and emotional tone. It’s starting to sound like someone from Taiwan instead of something that merely speaks Chinese. For brand voice, consumer communication and the creative industries, that’s a turning point. In customer service, content marketing, edtech and even counseling, an AI that talks like a person and feels warmer can greatly improve the user experience and make a business more competitive.

For example, companies can use GPT-4.5 to make customer service feel more personal, schools can use its emotional intelligence to support learning, and brands can use its creativity and copywriting skills to reach their target audience.

Wrap-up: GPT-4.5 is a gentle AI with opinions of its own

GPT-4.5 won’t change the world. Still, the direction it points to matters a lot to me: AI that is closer to people, takes a real part in communication, and expresses itself with more nuance. Content is everywhere, attention is scattered and creators are anxious, and what GPT-4.5 offers here is an early version of a new kind of partnership on top of a better productivity tool. For the Taiwanese market, which is eagerly adopting AI tools, GPT-4.5 could become an important virtual coworker at work and in daily life, and that may be the most interesting thing about this update. It isn’t perfect. Sometimes it rambles on its own, and its emotional responses can overshoot. But it’s learning how to be a partner in how we create, communicate and live, or at least an AI coworker that’s always online, never complains, and is fairly reliable.