OpenAI is reportedly working on an accelerated release for its newest model, Strawberry, slated to enhance the ChatGPT service. Initially expected to make its debut in the fall, Strawberry will now be available in the next couple of weeks.
Improved Reasoning and Analytical Skills
Originally named Q*, the model has been renamed ‘Strawberry.' The new AI is set to enhance OpenAI's present systems considerably. OpenAI presented ‘Strawberry' to U.S. national security officials during the summer, pointing to its potential applications in that field.
What sets Strawberry apart is its methodical approach—a “thinking” phase—before it outputs a response. Instead of delivering instant replies like its predecessors, Strawberry takes approximately 10 to 20 seconds to deliberate, aiming to produce more reasoned and insightful answers. The feature is expected to improve its performance in handling intricate math problems and unfamiliar questions.
The Strawberry model will be integrated into ChatGPT but will function independently within the platform. Users will likely have the option to select Strawberry from a range of AI models available in ChatGPT. The precise mechanism for this integration remains to be specified, but the model promises to bring distinct strengths and some differences in functionality compared to the existing ChatGPT service.
Userbase Expansion and Financial Goals
OpenAI has seen its user numbers swell, boasting over 1 million paying business customers, a significant rise from 600,000 in April. The figure includes users across its ChatGPT Enterprise, Team, and Edu offerings, launched at different times in 2023.
Concurrently, OpenAI is reportedly aiming to secure several billion dollars in a new funding round, with the company's valuation projected to surpass $100 billion. The development of the GPT-4 AI model alone has cost the company over $100 million, with further investments anticipated for future projects.
In internal trials, ‘Strawberry' demonstrated its prowess by solving the New York Times word puzzle “Connections.” An AI tested internally by OpenAI, likely ‘Strawberry,' scored over 90 percent on the MATH benchmark. Internal documents reveal plans to use ‘Strawberry' for autonomous internet searches, enabling it to conduct comprehensive research and planning.
The development approach for ‘Strawberry' is similar to Stanford's “Self-Taught Reasoner” (STaR) method, designed to boost AI reasoning skills. Former OpenAI chief researcher Ilya Sutskever, now running a startup focused on secure super AI, is credited with providing the foundational ideas for ‘Strawberry.'
