On Thursday, OpenAI released its latest innovation, GPT-4.5, which is hailed as the largest and most powerful model the company has developed to date. This new model is designed to be more versatile, moving beyond the STEM-focused capabilities of previous versions. According to OpenAI, GPT-4.5 excels in tasks such as writing, programming, and tackling practical problems, making it an invaluable tool for users across various industries.
OpenAI's CEO, Sam Altman, took to X to describe GPT-4.5 as the first AI model that truly feels like conversing with a thoughtful individual. In his post, he expressed his astonishment at the quality of advice generated by the AI, stating, "I have had several moments where I've sat back in my chair and been astonished at getting actually good advice from an AI." Altman also noted that while OpenAI aimed to launch GPT-4.5 for both the Plus and Pro subscription tiers simultaneously, they ran out of GPUs, a critical resource for powering the model. He assured users that the company would add tens of thousands of GPUs in the coming week to facilitate broader access.
Silicon Valley has been divided over the optimal methods for enhancing AI models. However, GPT-4.5 reinforces the belief that increased data and computational resources lead to superior performance. Mark Chen, OpenAI's chief research officer, confirmed to Big Technology that the company has not observed diminishing returns from scaling these models. "We're very rigorous about how we do this," Chen explained, emphasizing their predictive models based on prior training outcomes. He highlighted that while training costs are still high, OpenAI has identified cost-effective ways to operate larger models, citing significant reductions in inference costs since the debut of GPT-4.
OpenAI has made GPT-4.5 available in a research preview for users subscribed to ChatGPT Pro, which costs $200 per month, as well as developers utilizing the API. Next week, the company plans to extend access to ChatGPT Plus, Team, and Edu users. During a livestream demonstration, Amelia Glaese, a technical staff member at OpenAI, explained that GPT-4.5 represents the latest progression in OpenAI's unsupervised learning paradigm. This model emphasizes scaling up knowledge, intuition, and reducing instances of AI hallucinations.
In practical applications, GPT-4.5 stands out as the most adept conversationalist among OpenAI's models. Its improved ability to pick up on social cues provides a more intuitive and emotionally intelligent interaction compared to its predecessor, the STEM-focused reasoning model known as o1. Raphael Lopes, another member of OpenAI's technical team, demonstrated how GPT-4.5 could tactfully reframe an angry message to a friend, showcasing its emotional sensitivity. Lopes noted that GPT-4.5 is trained to be a better collaborator, resulting in warmer and more nuanced conversations.
OpenAI reported that GPT-4.5 outperforms its predecessors in terms of accuracy and significantly reduces the occurrence of hallucinations. The model's extensive knowledge base, improved alignment with user intent, and enhanced emotional intelligence make it particularly suitable for various tasks, including writing, programming, and solving practical problems, as stated in the GPT-4.5 System Card released by OpenAI on Thursday. This advancement positions GPT-4.5 as a game-changer in the realm of AI technology.
OpenAI has not yet responded to requests for further comments from Business Insider regarding the launch and capabilities of GPT-4.5, but the excitement surrounding this release suggests it will have a substantial impact on AI applications moving forward.
