OpenAI, the renowned AI research laboratory, is on the cusp of releasing two groundbreaking models, the Strawberry Project and Orion, which are anticipated to significantly enhance AI’s reasoning and generation abilities. According to AI engineer Tom Keldenich, these models are currently in development and could be launched as early as this autumn.
The Strawberry Project, as the name suggests, is designed to excel in complex reasoning tasks, particularly in mathematics and programming domains. It is also expected to bolster the language understanding capabilities of large models. In a test conducted by The New York Times, the Strawberry model demonstrated exceptional performance. This model’s ability to solve unseen mathematical problems and tackle tasks like The New York Times’ crossword puzzles, which are beyond the current AI’s capabilities, highlights its potential.
On the other hand, Orion is set to surpass GPT-4, the anticipated successor to the popular GPT-3 language model. What makes Orion unique is that its pre-training data is generated by the Strawberry model, indicating a synergy between the two that could surpass the capabilities of any current AI model.
The excitement surrounding the Strawberry Project stems from its positioning in the hierarchy of AI development. Analogous to the stages in autonomous vehicle development, OpenAI categorizes the pursuit of Artificial General Intelligence (AGI) into five levels. Strawberry occupies the second level, focusing on reasoning AI, where models can handle more intricate tasks with a broader skill set, approaching human proficiency. The subsequent levels involve autonomy, innovation, and organization.
Scheduled for a fall release, the Strawberry model could be integrated as an upgrade to ChatGPT, OpenAI’s popular conversational AI. It is reported that the company aims to launch the Strawberry model this autumn, with a simplified version for chatbots and a full-fledged model tasked with generating training data for the Orion project, addressing the scarcity of high-quality text on the internet.
OpenAI’s financials reveal that while the company currently earns $283 million per month from enterprise services and ChatGPT subscriptions, it remains in a net loss position. The successful launch of the Strawberry and Orion models is expected to fuel revenue growth and reinforce OpenAI’s position as an industry leader, especially amidst intensifying competition from rivals like Anthropic and Meta.
However, it’s worth noting that the enhanced reasoning capabilities of the Strawberry model come at the cost of response speed. This suggests that the model may be better suited for tasks that do not require immediate feedback, ruling out applications like GPT search, which necessitates rapid responses.
OpenAI’s recent updates, including the speed-oriented GPT-4, have set the stage for the Strawberry Project’s debut. With the promise of more accurate but potentially slower responses, the Strawberry model is poised to make a significant impact in sectors where precision outweighs speed.
In conclusion, the upcoming release of OpenAI’s Strawberry Project and Orion marks a crucial step in the progression towards AGI. As the AI community eagerly awaits these models, the potential they hold to revolutionize problem-solving, language understanding, and data generation is immense. OpenAI’s latest endeavor underscores its commitment to pushing the boundaries of AI technology and maintaining its status as a pioneer in the field.
【source】https://www.cls.cn/detail/1778853
Views: 1