OpenAI Launches New Model: ‘o1 Thinks Before Answering’
OpenAI has introduced its new generative AI model, “o1,” which is capable of thinking and answering highly complex questions, particularly in mathematics. The creator of ChatGPT aims to reduce the chances of hallucinations, which sometimes affect such AI programs.
OpenAI confirmed in an online statement that “o1 thinks before giving an answer.” The launch of this model represents a new step for the company toward achieving its goal of developing “general” artificial intelligence, meaning AI that resembles human intelligence.
Sam Altman, CEO of OpenAI, congratulated his team on this “new model: AI capable of complex and general thinking” via the “X” platform.
However, he admitted that this technology “is still imperfect, limited, and seems more impressive when used for the first time than after prolonged use.”
The company has made the beta version of “o1” available from Thursday, initially for ChatGPT users with a paid subscription.
Unlike its predecessors, “o1” takes more time and provides more detailed answers rather than generating them almost instantly.
Another significant difference is that the new model currently cannot process or generate any content other than text.
Investments
The launch of “o1” comes as OpenAI seeks to raise its valuation to around $150 billion, potentially making it one of the most valuable privately held companies globally, according to American media reports.
The list of potential investors includes major companies like Microsoft and Nvidia, the giant chip manufacturer. Other names mentioned in the media include Apple, which already uses the startup’s technology in its new generative AI system, Thrive Capital, a private equity firm, and MGX, a tech investment company founded by the UAE’s Council for Artificial Intelligence and Advanced Technology.
OpenAI’s release of ChatGPT at the end of 2022 triggered a race in Silicon Valley to develop generative AI programs capable of producing content based on simple everyday language queries.
Since then, all major tech groups—from Microsoft to Google and Meta—have been competing to provide tools designed to assist humans with everyday tasks, from writing messages to creative arts and education.
However, these AI-based assistance tools remain sophisticated machines that convincingly predict word sequences or pixel arrangements without actual thinking or understanding of the sentences or images they generate.
Thus, these models, trained using vast amounts of data from the internet, sometimes “hallucinate” by fabricating facts and inaccurate information, and they may struggle with solving simple mathematical problems.
Error Correction
OpenAI stated that it has tested its new model in fields such as solving mathematical problems and generating lines of code, confirming that “o1 competes with human experts in a wide range of benchmark tests requiring strong thinking skills.”
The company added that “o1” ranked “among the top 500 students” in a U.S. high school mathematics competition.
OpenAI explained that “o1 uses a chain of thoughts similar to a human who might think for a long time before answering a difficult question.”
The model “learns to recognize and correct its mistakes, breaks down complex steps into simpler ones, and tries a different approach when the initial one fails,” it added.
While the new model “hallucinates less” than its predecessor, “hallucinations cannot be said to be solved,” according to OpenAI researcher Jerry Touwrik, who was interviewed by the tech-specialized site The Verge. Another problem with the current models is that they function as black boxes to users.
The startup emphasized that the new model represents an improvement in terms of safety and alignment with human values because its reasoning is now “readable” and it better applies safety rules.
This article provides an in-depth look at OpenAI’s latest AI model, “o1,” its innovative capabilities, and the company’s strategic goals for advancing AI technology.