news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Cailianshe News, August 28 (Editor: Shi Zhengcheng)Earlier this month, Sam Altman, CEO of the tech startup OpenAI, posted a photo of his garden on the internet showing four tiny strawberries hanging gently from a branch.

The latest news on Tuesday showed that the picture contained important information:OpenAI's next major AI product, a new generation of reasoning model code-named "Strawberry Project", will be available as early as this fall.

(Source: X)

First of all, we need to explain why the industry is so excited about this "strawberry".

Similar to self-driving cars, OpenAI has divided the realization of general artificial intelligence (AGI) into five steps: the first is the lowest level we are currently at, conversational AI such as ChatGPT.

Then there is the second layer where "strawberries" are located - reasoning AI, where large models can handle more complex tasks, have more comprehensive capabilities, and are closer to human proficiency. The following layers are related to characteristics such as autonomy, innovation, and organization.

Available in autumn

U.S. media quoted people familiar with the project as saying that OpenAI's goal is to launch the reasoning AI code-named "Strawberry" as early as this fall. The project was previously known as Q*.

The specific product form may be part of the ChatGPT upgrade.The new AI will be able to solve "unseen math problems" and word puzzles published in the New York Times, tasks that existing AI cannot reliably perform.

In addition to solving math problems, the Strawberry Model can also be used for programming and answering a range of more subjective questions, including handling some multi-step tasks. OpenAI and its competitors hope that product upgrades will bring more revenue opportunities. The latest news revealed thatCurrently, OpenAI earns $283 million in revenue each month through enterprise services and ChatGPT subscriptions, but the company is still in a state of expanding losses.

It is reported that there are multiple versions of the "Strawberry Model". A scaled-down and simplified version is used for chatbots, while the full version of the model has a special mission -Generate training data for the next-generation flagship large model Orion to solve the problem of shortage of high-quality text on the Internet.

As its main competitors Anthropic and Meta are trying to catch up, OpenAI now needs an AI with explosive capabilities to re-prove its position as an "industry leader".

More precise, slower

As one of the few major updates of OpenAI at the end of this year, the company launched GPT-4o earlier this year, which is known for its speed. However, according to the latest report,The increase in the reasoning ability of the "Strawberry Model" will come at the expense of reaction speedTherefore, the main scenarios of this application should be those applications that do not require much feedback speed.

Therefore, it is not difficult to understand that scenarios such as GPT search that require instant and rapid response may not be suitable for using this project.