![Why was the field games restarted? [VIDEO]](https://thegeek.hu/wp-content/uploads/sites/2/2023/06/thegeek-fable-2023-trailer.jpg "Why was the field games restarted? [VIDEO]")
What is good at work? Instead of interlocking, we show real strategies, solutions and cases of use for a full day! Registration and details here!
GPT-4.5 has more advanced writing capabilities, more knowledge around the world and a more advanced personality than its predecessors. According to a document leaked by Openai, the new model improves more than ten times the efficiency of the GPT-4 account.
According to the previous news, Openai used O1 (symbol name: Strawberry) to train GPT-4.5 with artificial data.
According to the company, new supervision techniques were combined with traditional methods, such as soft -supervising soft learning (SFT) and learning to confirm on the basis of human comments (RLHF).
Despite its limits, GPT-4.5 significantly It produces fewer hallucinations, Like GPT-4O and a little less than the O1 model. According to the information that was leaked this week, the new model will only be available for Chatgpt Pro users, similar to the operator.
According to the information that appeared last week, Openai plans to launch GPT-4.5 and then GPT-5 by the end of May. SAM Altman, CEO of Openai, described GPT-5 as a system that integrates many technologies for the company, including the new O3 reasoning model, mentioned for the first time in December during Christmas announcements.
While O3-Mini was released last month, the full version of O3 It is scheduled to be published only as part of the upcoming GPT-5 system. This is in line with Openai's goal of creating large linguistic models to create a more popular system that can already be called artificial general intelligence (AGI).
Cover source: Cheng Sheen/Gety EM.
