When Openai began giving private demonstrations for the new GPT-4 technology at the end of 2022, its skills shock even the most experienced researchers in the AI. He could answer questions, write poetry and create a computer code in ways that looked far ahead of his time.
More than two years later, Openai released its successor: GPT-4.5. New technology means the end of an era. Openai said GPT-4.5 would be the latest version of the Chatbot system that did not make a “reasonable chain of thought”.
After this liberation, Openai’s technology can, like a human being, spend a significant time to think of a question before answering, instead of providing an immediate answer.
The GPT-4.5, which can be used for the power of the most accurate version of Chatgpt, is unlikely to create as enthusiasm as the GPT-4, to a large extent because AI research has shifted in new directions. The company also said that technology would “feel more natural” than previous Chatbot technologies.
“What the model puts is his ability to participate in warm, intuitive, naturally flowing conversations and we believe he has a stronger understanding of what users mean when they ask for something,” said Mia Glaese, Vice President of Openai.
In the autumn, the company introduced the technology called Openai O1, which was designed to justify through mathematics, coding and science duties. The new technology was part of a broader effort to build AI which can speak through complex duties. Companies like Google, Meta and Deepseek, a Chinese start, are developing similar technologies.
The goal is to build systems that can carefully and reasonably solve a problem through a series of distinct steps, each building in the last, similar to the way people. These technologies could be particularly useful for computer developers using AI systems to write code.
These reasoning systems are based on technologies such as GPT-4.5, which are called large language models or llms
LLMS learns their skills by analyzing huge amounts of text pounded by the internet, including Wikipedia articles, books and conversation records. By identifying patterns in all these texts, they learned to create text on their own.
To build reasoning systems, companies set LLMS through an additional process called aid learning. Through this process – which can be expanded in weeks or months – a system can learn behavior through extensive test and error.
By working through various mathematical problems, for example, he can find out which methods lead to the correct answer and which do not. If it repeats this process with a large number of problems, it can detect the standards.
Openai and others believe this is the future of AI growth. But in some ways, they have been forced in this direction because they have been exhausted by the internet data needed to train systems such as GPT-4.5.
Some reasoning systems go beyond ordinary LLMS in some standard tests. But standard tests are not always a good judge on how technologies will perform in real situations.
Experts point out that the new system of reasoning cannot necessarily be like a human being. And like other chatbot technologies, they can still get things wrong and do things – a phenomenon called illusion.
Openai said, since Thursday, the GPT-4.5 would be available to anyone who was registered with Chatgpt Pro, a $ 200-a-Month service that provides access to all the latest tools of the company.
(The New York Times sued Openai and her partner, Microsoft, in December for copyright violations of AI Systems.)
