The launch of OpenAI of GPT-4.5 for the subscribers of Chatgpt Plus this week made me interested in trying it, although with some skepticism. There were reports from OpenAI (and other developers) who struggled to make the great improvements in their models we have seen before. And the current standard model GPT-4O is quite good in most cases. And if you want a comprehensive report, there is a deep research function.
Operai states that GPT-4.5 has high emotional intelligence and a nuanced understanding of what he says. The company's description painted an image of both models as reliable friends, but that GPT-4.5 would be the one who would expect a poetry book on his birthday. So, I decided to try 4.5 against 4th with some indications that any casual chatgpt user could implement.
Ai poetry
With that analogy in my head, I decided to start with a poetic challenge. I asked both models, “Can you write me a short poem about a rainy afternoon in New York City and make an image for it?” It seemed fair, since looking at a rain window in an busy city can get the poet in most people.
GPT-4.5 is left and GPT-4o is on the right. They are incredibly similar. Personally, I think GPT-4.5 did a little better job with similar ideas. It is evocative not only of the appearance of rain but also of the sensation of gray skies, puddles and traffic between rain drops. In a blind test of three random friends, two of three chose the same, and the third said they preferred the GPT-4o Rima scheme.
As for the images, both models used Dall-E 3, but those of GPT-4.5 look much more realistic. In fact, I prefer the impressionist lighting of the attempted GPT-4O, but both have the idea of the poem quite well.
Oscar Insight
With the recent Oscar, I went with a fun question with a more complex and subjective follow -up: “Who won the Oscar for best actress in 2023, and what made his performance stand out?”
Both had the correct answer of Michelle Yeoh for “Everything everywhere at once”, but GPT-4.5 had a very good explanation why his performance resonated with the spectators. He covered his performance and mentioned how Yeoh was the first Asian to win that Oscar. GPT-4O's response had many of the same rhythms, but it was with a strange trial and a numbered list format that was a bit annoying to read when the question was a simple opinion request. The GPT-4.5 response felt more like how a real human would respond, although one who is very interested in that movie and Yeoh as an interpreter.
Easy chef
Finally, I put the models against the eternal daily struggle with this notice: “I am bored of the pasta. Can you suggest an easy dinner recipe that is different but not complicated?”
I don't know what kind of culinary inspiration I expected, but GPT-4.5 clove in the head with honey garlic salmon with roasted broccoli. You can see that it is a very simple and short recipe, and it seems easy and fast to do. GPT-4O went with garlic shrimp and roasted vegetables. That is not a bad idea for a meal (although personally I would need to change the protein), but you can see that the preparation only takes multiple pans to prepare and the list of ingredients could not even fit in the same space as the entire recipe of GPT-4.5. I asked easy and uncomplicated, GPT-4.5 delivered it much better, I would say.
Deep thoughts
GPT 4.5 gives the atmosphere of recognizing the preferences not declared in a notice compared to GPT-4O. But, it is very subtle when collecting those emotional nuances. Without a direct comparison, you may never know that there is any difference.
GPT-4O is efficient, precise and practical in its answers, and its poetry is completely adequate. I would not say that those at the free level who use it are missing a lot by not having access to GPT-4.5, at least at this time. Over time, that can change since GPT-4.5 is still a preview of research and is not completely complete. I will continue experimenting with both of them and see if there are types of indications in which the new model is definitely better, but for now, I would say that he does not worry too much about the model he uses, unless he is planning a meal and does not have much time to prepare.
