Is OpenAI Developing Mysterious 'Black Technology'? Discover the Secrets Behind OpenAI's Enigmatic Project 'Strawberry'

On the 12th, reports surfaced that OpenAI, the leader in generative AI, is working on a new, highly secretive AI model project called "Strawberry." This project appears to focus on enhancing the reasoning abilities of AI models. According to internal documents, OpenAI aims to improve its models' capacity to solve complex scientific and mathematical problems. The goal is for these models to not only generate answers to queries but also to plan independently and navigate the internet reliably for “deep research,” a level of functionality not yet achieved by current large language models.

OpenAI has chosen to keep details about “Strawberry” under wraps, with a spokesperson indicating that their goal is for AI models to understand and interpret the world as humans do. The understanding that the reasoning capabilities of AI will evolve over time is a common belief within the industry. The workings of “Strawberry” remain highly confidential, and no information regarding its release date has been disclosed.

Sources indicate that "Strawberry" originated from the Q algorithm model, designed to tackle challenging scientific and mathematical issues. Mastering mathematical capabilities is essential for the development of generative AI, as it significantly enhances reasoning abilities, potentially rivaling human intelligence—something current language models cannot yet achieve. The Q model was first mentioned in internal communications towards the end of last year, coinciding with the unceremonious departure of CEO Sam Altman from OpenAI due to controversies surrounding this project. Some insiders believe that Q may represent a significant breakthrough for OpenAI in its quest for artificial general intelligence (AGI), with its rapid development raising concerns about the implications for human safety.

Despite the controversies surrounding OpenAI's leadership, the company's ambition to enhance the reasoning capabilities of its AI models through the "Strawberry" project is evident. Altman has previously emphasized that the future of AI development will hinge on improving reasoning skills. An internal presentation showcased a research project purportedly demonstrating human-like reasoning abilities, although it remains unclear whether this project is directly linked to “Strawberry.”

Reports suggest that “Strawberry” employs a specialized “post-training” method, refining models after extensive pre-training on large datasets to enhance performance on specific tasks. This approach mirrors Stanford University's 2022 development of the Self-Taught Reasoner (STaR), which enables AI models to create their own training data, theoretically allowing them to reach higher levels of intelligence.

OpenAI's recently unveiled five-level roadmap for AI development outlines the anticipated evolution of AI capabilities. Currently, models are categorized at the first level—basic chatbots—while expectations project an imminent transition to the second level, featuring human-level reasoning.

Furthermore, OpenAI is focusing on enhancing the ability of large language models to execute long-horizon tasks that require advanced planning over extended periods. To achieve this, OpenAI envisions “Strawberry” facilitating the creation, training, and evaluation of models capable of deep research, utilizing computer agents for autonomous web browsing and decision-making based on their discoveries.

If successful, the "Strawberry" project could redefine AI capabilities, leading to significant scientific advancements, the development of new software applications, and the ability to undertake complex tasks independently—bringing humanity one step closer to realizing artificial general intelligence.