news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

cailianshe news, september 11 (editor: shi zhengcheng)openai, the global ai leader that has been silent for several months, seems to be about to make a big move.

according to a report by the information on tuesday, citing multiple people familiar with the matter,openai plans to officially launch the "strawberry" ai model with reasoning capability as its core in about two weeks and add the chatgpt service.due to the vigorous promotion of "strawberry" by openai ceo altman and others, and the fact that the company has not had any exciting news for a while, the outside world has certain expectations for this new product.

a large model that actively "meditates"

the report said,as a large model with reasoning as its selling point, strawberry will be more powerful and easier to use than the current gpt-4o when facing complex, multi-step questions.。

when using gpt-4o, if you want to ask a multi-step reasoning question, in order to avoid the machine answering randomly, the user often needs to clearly tell the robot every reasoning step when answering the question - similar to entering the "thinking chain prompt word". when using "strawberry", it is not so troublesome, and the output result will be better. for example, this model can solve the new york times crossword puzzle game.

(source: the new york times)

therefore, in addition to mathematics and code, strawberry is also suitable for answering broader and more subjective questions. for example, if you ask it to "brainstorm" a marketing strategy, the answer given by strawberry is closer to the actual scenario of the user and pays more attention to details, such as generating an execution plan in weeks.

the core reason for the difference in answer quality isthe difference is that the "strawberry" model will not answer users' questions immediately, but will "think" and ask users additional questions to provide more comprehensive answers.

of course, this step is also a potential weakness of "strawberry" as an internet product - the report said,it may take 10-20 seconds for the model to start answering the questionfor industries that increasingly emphasize real-time communication with ai,whether paying users can accept an interaction interval of more than ten seconds is a question。

the more fatal problem is that when answering questions that gpt-4o can also reason about (in fact, humans still don’t know where the limits of ai’s capabilities are), strawberry may think for a while and then only give a slightly better answer. people familiar with the matter who have experienced this difference said,the improvement in results is not worth waiting so long, gpt-4o, which can speak fluently, provides a better experience.

the insider also said:in internal testing, strawberry also had problems remembering conversation history.at the same time, this model can only input and output text and cannot process images, and has not yet crossed the critical multimodal threshold.

business strategy is also a highlight

even if the revelations are accurate, the highlight of this matter, in addition to the capabilities of strawberry itself, is openai’s considerations on business strategy.

according to earlier reports,in internal discussions, openai is considering setting a higher price for the upcoming "strawberry" large model, which specializes in reasoning, or the next-generation flagship model orion.a more practical approach is that openai may add a new paid tier in addition to the existing chatgpt plus, providing unlimited access to strawberry or improved response speed. of course, all this is premised on a large number of users recognizing the improvement of strawberry's capabilities.

under the current pricing system, chatgpt plus for individual users costs $20 per month, and the charges for enterprise users range from $25 to $60 per month per account.

last month, there was news that openai could earn $283 million in revenue per month through enterprise services and chatgpt subscriptions, but the company was still in a net loss state.