Highlights:

  • AGI is a sophisticated type of AI that can understand, learn, and apply knowledge across a broad range of tasks, resembling human cognitive abilities.
  • First reported by The Information, the new model, known internally as ‘Strawberry,’ is said to solve unfamiliar math problems, handle high-level tasks like creating marketing strategies, and tackle complex word puzzles.

Reports suggest that OpenAI is ready to unveil a new AI product with advanced capabilities designed to tackle problems and tasks that existing AI models can’t handle.

According to an initial report by The Information, the new model, internally known as ‘Strawberry,’ is claimed to solve unfamiliar math problems, handle advanced tasks like creating marketing strategies, and tackle intricate word puzzles. For instance, it successfully solved the New York Times word puzzle ‘Connections.’

Previous reports suggest that the model scores over 90% on the MATH benchmark, which features high-level math problems. In contrast, GPT-4 scored just 53% and GPT-4o achieved 76.6%. As of July, GPT-4o held the highest MATH benchmark score among available AI models, so if Strawberry performs as expected, it will place OpenAI significantly ahead of its competitors.

The development history of the new model adds an intriguing layer to the news. Although the name ‘Strawberry’ may seem pleasant and uncontroversial, the model was previously known internally as Q* (pronounced Q-Star). It played a significant role during the brief turmoil at OpenAI last year, which included the removal and subsequent reinstatement of CEO Sam Altman.

In November, reports suggested that Sam Altman’s removal was influenced by worries over a major AI breakthrough at the company. Prior to his ouster, a group of OpenAI researchers sent a letter to the board raising concerns about the potential risks of advanced AI, particularly regarding Q*, the model now called Strawberry.

At the time, it was claimed that Q* could represent a significant advancement in the quest for artificial general intelligence (AGI). AGI is a sophisticated type of AI capable of understanding, learning, and applying knowledge across various tasks, akin to human cognitive abilities.

The potential risk of AGI revolves around losing control over AI systems, which could lead to unintended consequences and the development of goals that conflict with human values, potentially causing significant harm. In simpler terms, there’s a concern that an AGI model could end up like Skynet from the ‘Terminator’ movies.

The Information reports that Strawberry might be released sometime this fall.