The rumors about Strawberrythe next artificial intelligence of OpenAI focused on reasoning, have emerged on several occasions over the past year, but are about to come to an end. The Information claims that the company led by Sam Altman plans to launch it within two weeks and has revealed some new details about this project that, for the most part, remains shrouded in mystery.
According to the outlet, Strawberry is different from other language models and, Instead of responding immediately to a request, he will take some time to ‘think’ about the response.. This would be between 10 and 20 secondsduring which it will search for the information it needs on the web, and will help prevent AI from fabricating answers and giving incorrect information, a common problem in current language models.
It will be part of the ChatGPT servicebut does not specify how. It could be another option for paying subscribers of ChatGPT Pluswho can currently choose between GPT-4o, GPT-4o Mini and GPT-4 language models in the chatbot, or a standalone offering that would require a new subscription.
The new AI, moreover, It would not be multimodal and could only understand and generate text, but not work with images, unlike GPT-4o.
The details published by the outlet are very scarce and some differ from the information about Strawberry advanced by Reuters last July. At the time, it claimed that Strawberry had significantly more advanced reasoning capabilities, but it would not be so much a new independent language model, but rather a a reasoning technology that improves other modelsWith Strawberry, an artificial intelligence will be able to plan ahead and surf the internet independentlywithout human intervention, to carry out what the company calls ‘deep research’.
What does it mean that Strawberry ‘thinks’?
This is the big question that has no clear answer, but suggests higher quality and fewer errors in the responses generated by ChatGPT. Shortly before Reuters’ report on Strawberry, OpenAI explained the 5 levels what is AI until reaching general artificial intelligence or AGI. The level 1 These are the chatbots we have known since OpenAI launched ChatGPT almost two years ago. They predict which word is most likely to follow another when generating a response, and for this reason they can give wrong answers, since they do not really understand what is being said. Level 2 are the so-called ‘reasoners’capable of solving problems at a human level, and This is where Strawberry would fit in.
According to the media, OpenAI had an AI capable of achieving in July a 90% score on MATH testsThese are a set of mathematical problems used as a reference to evaluate the capabilities of artificial intelligence models in solving advanced level mathematics problems. This ability is what allows it to be compared to a human..
The new AI first came to light almost a year ago, under a different name, when Sam Altman was briefly fired by OpenAI’s board of directors for loss of trust. As Reuters reported at the time, One of the reasons was the development of Q*now Strawberry, which he referred to as a powerful artificial intelligence that could threaten humanityd.
The fear of what AI can do has been tempered in public discussion almost two years after the launch of ChatGPTas more applications are released, they are integrated with more services and users become more familiar with them. If the timeframes advanced by The Information hold true, we will soon know exactly what Strawberry is and what it can do.