Although the improvements are as gradual as its name suggests, GPT-4.5 is still the most ambitious version of OpenAi so far. Run at the end of February as a research progress, which basically means that Openai considers it a beta version, GPT-4.5 uses more calculation power than its previous models and has trained with more data.
What is the scope of GPT-4.5? Who knows, because developers do not say it. And where do this additional training data come from? They don’t say anything about it either. Borrowing a phrase from the successful Apple TV program, Dismissal, OpenAI is presenting the supposed improvements of this new model as mysterious and important.
When compared the reference tests of the competition models and the “reasoning” versions of OpenAi, the advantages of using GPT-4.5 are not clear at first glance. However, in the model system tab and in a previous interview with Wired, OpenAi researchers who worked in GPT-4.5 affirmed that the improvements are noticed in the anthropomorphic aspects of the model, such as a stronger intuition and a deeper understanding of emotions. After sitting at the OpenAI office last year and listen to managers to talk about the plan of the Startup to continue producing chatgpt as software Useful, this was not the launch I expected by 2025. Instead of adopting a more functional approach, this model tries to be more emotional.
Openai has not stopped increasing its number of contracts with companies, so it would be expected that the company would launch large -scale versions with incorporated practical applications, especially in the most expensive and powerful version of its Chatbot. However, GPT-4.5 further adjusts to the results of an academic research group that has put all its efforts to pursue general artificial intelligence (AGI), a theoretical version of the algorithm skilled enough to replace white-collar workers and practically divine in their ability to process information.
Although Openai would argue that these two ways are intertwined and are equally important, if your short -term goal is to make money with Chatgpt, the pharragoso launch last week makes no sense; It is super expensive and offers marginal profits that only the ever -time users of the Chatbot They can notice. But if your global mission is to build a beneficial AGI, which is still The main objective of OpenAIthen imitating the nuances of human emotions and soft skills is a critical area of improvement. It is there where the company could maintain its leadership position while other competitors in the generative AI career, such as Depseek’s R1 model, much cheaper, advance in other innovations.
How to access GPT-4.5
As with most of the new functions and models that arrive for chatgpt, OpenAI’s payment subscribers will be the first to access GPT-4.5. In this case, Openai is unlocking the first access for Chatgpt Pro subscribers who pay the bulky $ 200 per month. The great deployment of GPT-4.5 to the other payment levels (Plus, Team, Enterprise and Edu) will occur during this week and the next. The previous models have also ended up reaching the free version of Chatgpt, but the company is not planned to launch GPT-4.5 to all users, due to their size and computer requirements.
When available in your account, GPT-4.5 will be one of the many nested options in the models drop-down menu that appears when doing click In the word Chatgpt, at the top of the screen. In my pro account, this has raised the total number of models available to the whopping of nine different options among which I now have to choose. Openai developers have told me that they hope to significantly simplify this process in the future and make the AI tool choose the most appropriate model for each question that the user writes or pronounces.
First impressions of the new OpenAI model
The provisional holder that I put in this article was “with GPT-4.5, OpenAi is lost in the AGI sauce”. And although no headline that includes the name of the model will seem poetic, that is a bit problematic. Writing overwhelming and concise headlines is a difficult ability that requires clear communication and a certain aesthetic taste, it often implies the intervention of several editors before transmitting the perfect message. I was curious to know if Chatgpt would be able to punch that headline, so I tried using both the most recent model and GPT-4O, an earlier version that the company describes as “great for most tasks.”
Among all intangible improvements, GPT-4.5 was much more capable of writing a convincing holder. The results of GPT-4O were less interesting and had less variety in general, with the exception of this meaningless bombing: “With GPT-4.5, OpenAi maintains a foot in the future and another in the Chatbot“. Here is a much better shot by the new model:” With GPT-4.5, OpenAi stumbles with their own ambitions. “It is quite similar to the original, but potentially clearer to readers. After a certain dismay, Wired’s human editors decided to opt for the headline generated by GPT-4.5. It seems fair.
The OpenAi CEO has said that GPT-5 will be available for all chatgpt subscribers.
Changing the subject, I asked why the price of a dozen eggs is going up even more during the start of Trump’s presidency than under Biden, mainly to see which model would be most successful in the analysis of articles web On a political issue. The differences here were more subtle, but GPT-4O seemed prone to sermolize and repeat, while GPT-4.5 did a better job understanding my intention and representing succinctly multiple points of view.
And the AI as a research partner? Right now I’m working on a report and wanted to see if the tool Deep research of chatgptin which you think for a little more time and produce deeper answers, it would be really useful during the preparation process. After introducing notes in GPT-4.5 and GPT-4o and wait a few minutes, both results disappointed me equally. While it is true that the long format can be useful for those who do not have experience as reporters, the questions I suggested asking my sources for the report did not surprise me or liked. I took better advantage of the time again reading my written notes instead of the reports of 4,000 words of Chatgpt.
After spending a couple of days chatting in general with GPT-4.5, the most positive thing I experienced was that the model makes Chatgpt look less annoying and more conversational. This may sound trite, but it is an important factor of the user experience. Small details such as the use of casual language and single -word phrases made the Bot It will be less cloying during our chats. I see myself choosing GPT-4.5 so as not to have the feeling of asking for help from an information flatter.
#GPT4.5 #OpenAI #stumbles #ambitions
