The Future of AI: OpenAI's “Strawberry” Model

OpenAI's much-anticipated “Strawberry” model is on the verge of being unveiled. A report from The Information indicates that as OpenAI continues to secure additional funding, its researchers are gearing up to introduce a groundbreaking AI product. This new model is anticipated to outperform current AI systems in tackling intricate challenges. Insiders familiar with the project have disclosed that this AI, internally dubbed “Strawberry” (formerly known as Q*, pronounced “Q Star”), will be integrated into ChatGPT's functionalities this fall.

“Strawberry” is engineered to solve complex mathematical problems, a feat that current chatbots struggle with, and it has also been programmed to handle programming-related queries. However, its abilities are not limited to technical challenges. Given additional “thinking” time, it can engage with customer inquiries on more subjective matters, such as marketing strategies. OpenAI employees have showcased its linguistic capabilities, illustrating how it can adeptly solve intricate word puzzles like the New York Times’ “Connections” puzzle.

The report emphasizes that this technology is crucial for future AI products, especially those aimed at addressing multi-step tasks. OpenAI and its competitors are actively exploring these intelligent agents as potential avenues for revenue growth. The company's business is witnessing rapid expansion. Although monthly expenses may outpace revenue, earnings from the sale of large language models and ChatGPT subscription services have surged nearly threefold compared to last year, now totaling $283 million. Presently, OpenAI's valuation stands at $86 billion. However, the report notes that the company's future is somewhat contingent on the new flagship large language model currently under development, codenamed Orion.

This model is intended to enhance the capabilities of the existing flagship large language model, GPT-4, which was launched earlier last year. Competitors have already introduced large language models that rival GPT-4's performance.

Uncertainty surrounds the potential release of a chatbot version of “Strawberry” this year, a simplified iteration known as the distilled version, which aims to deliver high performance while being more user-friendly and cost-effective. Insiders suggest that OpenAI is also utilizing a larger variant of “Strawberry” to produce data for training Orion. This AI-generated data, referred to as “synthetic data,” indicates that “Strawberry” could assist OpenAI in overcoming the challenges associated with sourcing high-quality training data from the internet.

Moreover, insiders believe that “Strawberry” could help diminish hallucinations in upcoming OpenAI agents. Alex Graveley, CEO of Minion AI and former Chief Architect of GitHub Copilot, proposed that leveraging “Strawberry” to generate superior training data could aid OpenAI in curtailing hallucinations in its models. Graveley noted, “Imagine a model devoid of hallucinations; you pose a logical puzzle, and it resolves it correctly on the first attempt. This could be achievable due to the reduced ambiguity in the training data, leading to fewer errors.”

Earlier this month, OpenAI CEO Sam Altman stirred speculation on social media by posting an image of a strawberry without further context. Insiders revealed that this summer, OpenAI presented the “Strawberry” model to U.S. national security officials. “We believe we now possess sufficient data for the model,” Altman remarked at a May event, likely alluding to Orion. “We’ve been executing various experiments, including generating synthetic data.”

Currently, Altman is also working to secure additional funding for the company and identify strategies to mitigate losses. Sources indicate that since 2019, OpenAI has raised approximately $13 billion from Microsoft as part of their collaboration, with the contract extending until 2030. Additionally, insiders mentioned that the terms of collaboration may evolve, affecting how OpenAI finances its cloud server rentals, which represent the company's largest expense. As of the report's publication, OpenAI's spokesperson had not provided a comment.

The potential of “Strawberry” in addressing the naming and reasoning abilities has been previously reported by Reuters. The industry largely views an AI capable of solving complex mathematical issues as a lucrative application area, especially given that existing AI systems often falter in math-heavy sectors like aerospace and structural engineering. Many AI researchers have observed that conversational AIs, such as ChatGPT and its equivalents, frequently provide incorrect responses, evidenced by a recent viral challenge where chatbots struggled to compare the numbers 9.11 and 9.8.

Furthermore, advancements in mathematical reasoning could enhance AI models' performances in managing dialogue queries, such as customer service inquiries. Companies like Google and various AI startups are also actively developing reasoning technologies. Last month, Google DeepMind announced that its AI could outperform most human participants in the International Mathematical Olympiad.

OpenAI's competitor, Anthropic, previously indicated that its latest large language model, bolstered by enhanced reasoning capabilities, can produce more complex software code and respond to inquiries regarding charts and graphs.

To improve models' reasoning skills, several AI startups have adopted a "shortcut" methodology, breaking down problems into smaller components, though this approach can be slow and costly.

Despite the uncertainty regarding the release of “Strawberry,” expectations for Orion remain high, as OpenAI aspires to maintain its competitive edge and continue its impressive revenue growth.

Interestingly, the development of “Strawberry” is also tied to former OpenAI Chief Scientist Ilya Sutskever. Reports suggest that a few years ago, Sutskever initiated research on the “Strawberry” model. However, Ilya has recently announced his departure from OpenAI to launch a new venture called Safe SuperIntelligence (SSI). Prior to his exit, OpenAI researchers Jakub Pachocki and Szymon Sidor built upon Sutskever’s work to create a new mathematical solving model named Q*, raising concerns among researchers focused on AI safety.

During the Q* development phase, OpenAI researchers also introduced a variant concept dubbed “test-time computation,” aimed at enhancing the problem-solving abilities of large language models. This method allows the model more time to contemplate all aspects of a user command or inquiry. At that time, Sutskever also published a blog post related to this research.

Looking ahead, OpenAI is determined to expedite the release of “Strawberry” to sustain its leadership in the AI sector and navigate the intense competition from other firms. Meanwhile, Orion, as OpenAI’s next pivotal large language model, carries significant responsibility for the company’s future trajectory. Its performance will substantially influence OpenAI's long-term outlook.

As OpenAI and similar companies continue to achieve breakthroughs in the AI domain, there is an increasing emphasis on safety considerations within the industry. As AI technology gradually evolves towards more sophisticated reasoning capabilities, the development of “Strawberry” and Q* has ignited extensive discussions regarding AI ethics and safety.

The challenges confronting OpenAI and the broader AI industry encompass not only technological advancements but also the need to balance innovation with safety. In the future, as more AI products like “Strawberry” come to market, the trajectory of AI development will diversify, impacting various facets of society.

Ultimately, the introduction of “Strawberry” will signify a significant leap forward in AI reasoning capabilities for OpenAI. Whether it enhances the existing GPT-4 model or supports the forthcoming Orion model, “Strawberry” is set to play a crucial role.

OPENAI'S STRAWBERRY: Groundbreaking NEW Details FINALLY Revealed! - Explore the details behind OpenAI's upcoming "Strawberry" model and what it means for the future of AI.

Chapter 2: Orion: OpenAI's Next Flagship Model

As OpenAI continues to innovate, its next major model, Orion, is generating significant interest. This model is designed to build upon the successes of GPT-4 and address the challenges faced by existing AI systems.

OpenAI's Next Big Model is Called Orion - Discover what we know about OpenAI's Orion model and its potential impact on AI technology.