OpenAI, creator of the famous ChatGPT, is working on an innovative approach to its AI models as part of a project code-named “Strawberry“. This news, which had not been disclosed until now, comes at a crucial time for the Microsoft-backed startup, which is committed to proving that its models are able to provide advanced reasoning capabilities, an ambitious goal that could revolutionize the sector.
OpenAI Strawberry: AI model that thinks like a human
According to internal sources and documents reviewed by Reuters, several OpenAI teams are working on Strawberry, a project shrouded in mystery even within the company itself. The goal is to enable artificial intelligence language models of OpenAI not only to generate answers to queries, but also to plan ahead, independently browsing the Internet to perform what the company defines as “in-depth research".
This is a complex challenge which, as emerges from interviews with numerous industry experts, has so far eluded existing AI models.
How Strawberry works remains an enigma, but its potential is undeniable. If OpenAI succeeds in drastically improving the reasoning capabilities of its models, the repercussions could be enormous. Imagine gods AI models capable of making great scientific discoveries, plan and develop new software applications, or tackle complex problems with intuition and logic of a human being.
The prospects are fascinating and disturbing at the same time, raising questions ethical and philosophical on the role of artificial intelligence in our society.
But the road to this goal is still long and full of obstacles. Large language models like those developed by OpenAI excel at synthesizing text and composing elegant, but often, prose they fail when faced with seemingly trivial problems for the human intellect, such as recognizing logical fallacies or playing tic-tac-toe. When encountering these situations, models tend to “hallucinate” false information, demonstrating the limitations of current AI.
It should be remembered that they are called linguistic models for a reason: reasoning is not the core of their functioning and it is normal for them to have limits. However, OpenAI with “Strawberry” aims to fill this large gap with the ultimate goal of create something truly majestic.
