OpenAI, known for the popular ChatGPT, has focused its efforts on a new secret project codenamed Strawberry, set to surpass traditional neural networks that can only answer questions. At its core is a system akin to artificial general intelligence (AGI), capable of solving tasks at a human level. At the company's latest internal meeting, a demo version was showcased, demonstrating the new AI model's human-level reasoning capabilities. Following this, OpenAI representatives cautiously hinted at the impending official launch of the technology. Details of the project remain confidential.
The primary goal of Strawberry is to create a neural network that can not only respond to questions but also autonomously conduct online research for scientific purposes. This approach would allow OpenAI's artificial intelligence to solve tasks at a higher level, planning and executing complex actions over an extended period.
Experts assert that reasoning is foundational for achieving AI capabilities that surpass human intelligence. Language models that exist today and are actively used skillfully generate and generalize texts across various domains. However, they often falter at solving logical tasks that humans handle intuitively, leading to errors and so-called "hallucinations."
Initially named Project Q last year and considered groundbreaking, the new model could already solve complex scientific and mathematical problems that other systems could not handle. This was achieved through further training after initial data processing. Meanwhile, there was significant concern among OpenAI employees about the lack of proper safety measures for the development and commercialization of such models. Plans to wind down work on Q led to the dismissal of Sam Altman, who ambitiously aimed to create a high-powered AI and stated that the most significant progress in artificial intelligence would be linked to the latter's ability to reason.