Image: © Photocreo Bednarek/Stock.adobe.com
The general-purpose model does not think before it responds, said OpenAI.
OpenAI has launched a research preview of GPT-4.5, its “largest and best model for chat yet”. The large language model (LLM) has a broader training base, an improved ability to understand user intent and a greater ‘EQ’, said OpenAI, resulting in a more “natural” sounding model.
According to the AI start-up, the new model has a “better understanding of what humans mean and interprets subtle cues or implicit expectations with greater nuance”, while also showcasing “stronger aesthetic intuition and creativity”.
Although, it doesn’t think before it responds, OpenAI wrote in its launch blog yesterday (27 February), and suggested that the model is more “general-purpose”, unlike OpenAI’s o1 and o3-mini, which are reasoning models.
While on X, Sam Altman, OpenAI’s CEO said that GPT-4.5 is the “first model that feels like talking to a thoughtful person”.
“I have had several moments where I’ve sat back in my chair and been astonished at getting actually good advice from an AI,” he said. “This isn’t a reasoning model and won’t crush benchmarks. It’s a different kind of intelligence and there’s a magic to it I haven’t felt before.”
Some of the use cases suggested by OpenAI include asking GPT-4.5 to help you through a tough time. According to the start-up, while GPT-4o would give factual advice to help you address your concerns, GPT-4.5 produces a more human-like, natural response.
GPT-4.5 was trained on Microsoft’s Azure AI supercomputers, which produced a model with a “deeper understanding of the world”, said OpenAI. The result is a model with considerably lower hallucinations at just over 37pc when compared to o1 at 44pc, and increased accuracy of 62.5pc while o1 rests at 47pc.
GPT-4.5 is rolling out to software developers and ChatGPT Pro users, while other paid users will gain access to the model in the next few weeks.
OpenAI debuted o1, the LLM focused on “complex reasoning” last September. The model is capable of “thinking” for a longer time when solving a problem and was a dramatic improvement on previous models’ ability to reason and answer more complex questions.
While the start-up released GPT-4 all the way back in 2023, which, according to OpenAI, was a marked improvement from its predecessor GPT-3.5, being “40pc more likely to produce factual responses than GPT-3.5”.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
