When Openai began giving private demonstrations of her new GPT-4 technology at the end of 2022, her skills shook even his most experienced researchers. It can answer the question, write poetry, and generate computer code in ways that looked long before its time.
More than two years later, Openai has released her successor: GPT-4.5. New technology means the end of an era. Openai said that GPT-4.5 would be the last version of his chatbot system that did not “justify the chain thinking”.
After this release, Openai’s technology, as a man, can spend a considerable time thinking about a question before answering, rather than providing an immediate response.
GPT-4.5, which can be used to strengthen the most expensive version of the chatgt, is unlikely to generate as much excitement as possible in the GPT-4, in large part because the research of it has moved in new directions. However, the company said technology would “feel more natural” than its previous chatbot technologies.
“What separates the model is his ability to engage in warm, intuitive conversations, of course, and we think it has a stronger understanding of what the user means when looking for something,” said Mia Glaese, Vice -President of Openai.
In the fall, the company introduced the technology called Openai O1, which was created to reason through tasks that include mathematics, coding and science. New technology was part of a broader effort to build what can reason through complex tasks. Companies like Google, Meta and Deepseek, a Chinese start, are developing similar technologies.
The goal is to build systems that can carefully and logically solve a problem through a series of discrete steps, each building in the latter, similar to how people reason. These technologies can be particularly useful for computer programmers using systems to write code.
These reasoning systems are based on technologies such as GPT-4.5, which are called large language models, or LLM
LLM learn their skills by analyzing large quantities of text drawn from the entire internet, including articles, books and Wikipedia conversations. By marking patterns throughout that text, they learned to generate the text themselves.
To build reasoning systems, companies set LLM through an additional process called reinforcement lesson. Through this process – which can extend by weeks or months – a system can learn behavior through test and extensive error.
Working through various mathematics problems, for example, can learn which methods lead to the right answer and which do not. If you repeat this process with a large number of problems, it can identify the models.
Openai and others believe this is the future of developing it. But in a way, they have been forced in this regard because they have completed the internet data needed to train systems like GPT-4.5.
Some reasoning systems exceed the usual LLM in certain standardized tests. But standardized tests are not always a good judge of how technologies will perform in real world situations.
Experts show that the new reasoning system cannot necessarily reason as a man. And like other chatbot technologies, they can still mistake things and create things – a phenomenon called hallucination.
Openai said that, starting on Thursday, the GPT-4.5 would be available to anyone who agreed on Chatgpt Pro, a $ 200 service per month that provides access to all the latest funds of the company.
(The New York Times sued Openai and his partner, Microsoft, in December for violation of copyright content of news content.)