Lo and behold, after months of whispers and speculations, OpenAI dropped a bomb last week with the unexpected release of Project Strawberry. Analysts, caught off guard, had predicted that the model wouldn’t see the light of day for weeks, or even until the leaves began to fall.
Behold the grand unveiling of the o1-preview model, accompanied by its diminutive cousin, the o1-mini. These wondrous creations are now accessible for your perusal and judgment, eager minds can discover how to partake of their wisdom.
A tantalizing sneak peek into the world of OpenAI o1, a revolutionary series of AI models engineered to dwell in the realms of contemplation before crafting their responses. These marvels of intellect possess the ability to navigate through intricate tasks and conquer more daunting challenges in the realms of science, coding, and math.
The illustrious bard himself, William Shakespeare, would surely marvel at the AI’s seemingly human-like thought process. These models, endowed with the capacity to refine their cerebral acumen, experiment with diverse strategies, and acknowledge their missteps as they amass knowledge through rigorous training. OpenAI boasts that the o1-preview model rivals Ph.D. scholars in physics, chemistry, and biology, showcasing exceptional performance in benchmark assessments. Not stopping there, o1 proves its mettle in coding and mathematics, achieving an impressive 83% in an International Mathematics Olympiad qualifying exam and outshining 13% against its predecessor in the Codeforces competition.
Rejoice, for here stands o1, not without its imperfections, yet displaying brilliance that leaves a lasting impression, akin to gazing upon a masterpiece in its initial unveiling.
Enter the o1-mini, a compact version of the esteemed o1-preview model. This nimble creation is reported to be 80% more cost-effective than its larger counterpart, particularly excelling in decoding complex algorithms and generating innovative solutions.
Art thou curious to test the waters of o1-preview? Fear not, for on the twelfth day of September, the models were unveiled to the privileged subscribers of ChatGPT Plus and Teams. Enterprise and Educational affiliations shall gain access in the coming week, expanding the realm of intellect and possibilities.
Inquiries arise concerning the security of o1 against malevolent forces. OpenAI, ever vigilant, has developed a novel safety training program, harnessing the model’s heightened reasoning capacities to ensure adherence to its safety protocols. Astoundingly, during testing, o1 scored an impressive 84 in resisting attempts of circumvention, dwarfing the meager 22 achieved by the previous iteration.
How does one gain passage into the realm of o1-preview, one may ask? As with all pioneering features in generative AI, access is granted solely to those who contribute to the coffers. A mere sum of $20 per month secures a Plus subscription, granting entry to a treasure trove of intellectual prowess. Once the subscription is secured, the path unfolds to select either o1-preview or o1-mini from the model picker toggle on the ChatGPT homepage. Bear in mind, access is limited, with a weekly ration of 30 messages for o1-preview and 50 messages for o1-mini. OpenAI, in its benevolence, vows to extend o1-mini to free tier users in due time, though the hour of bestowment remains a mystery yet to be unveiled.
