On August 13, ChatGPT posted on social media X that the GPT-4o model has been updated again, and it is emphasized that this time it will be very different.This model is not the latest API GPT-4o 0806 model of OpenAI. Currently, free users can also use this model, but there is a limit on the number of times it can be used.

Before the official announcement, there were rumors from foreign media that OpenAI would soon announce the "Strawberry" project in an update to ChatGPT.These speculations are not groundless.

ChatGPT had earlier posted a strawberry emoticon on X, which seemed to be hinting at something.

(Source: X)

What exactly is the "Strawberry" project?

Last month, OpenAI’s internal team was revealed to be developing the “Strawberry” project, which aims to enhance the reasoning capabilities of OpenAI’s models and its ability to handle complex scientific and mathematical problems, so that large models can not only generate query answers, but also plan ahead. , in order to browse the Internet autonomously and reliably and conduct "in-depth research" defined by OpenAI.

The ability to browse the Internet autonomously is critical to this vision. By giving artificial intelligence the ability to independently explore the web, Strawberry can instantly access the latest information, explore a variety of sources and perspectives, and continuously expand its knowledge base. This ability can be very valuable in fields where information is evolving rapidly, such as scientific research or current affairs analysis.

The predecessor of this "Strawberry" project was the Q* algorithm model. Q* can solve difficult scientific and mathematical problems. Mathematics is the basis for the development of generative AI. If the AI ​​model masters mathematical abilities, it will have stronger reasoning capabilities and even rival human intelligence. This is something that current large language models cannot yet achieve.

For example, many current large AI models are not so sensitive to numerical calculations. When you ask how many r's are in the word strawberries, gpt gives the hilarious answer of only two r's.

(Source: GPT4.0)

The potential applications of this advanced artificial intelligence model are vast and exciting. These include:●Science research: Speeds up literature reviews and helps generate hypotheses

●Business Intelligence: Provides instant market analysis by synthesizing large amounts of data

●Educational Learning: Create a personalized learning experience through in-depth, up-to-date content

●Software development: Assist in completing complex coding tasks and solving problems

Altman once emphasized that the key to future AI development is reasoning ability

According to people familiar with the matter, the "Strawberry" project includes a specialized "post-training" method, that is, after the generative AI model has been pre-trained on a large data set, the model is further adjusted to improve its performance on specific tasks. This is similar to the "Self-Taught Reasoner" (STaR) method developed by Stanford University in 2022.

Noah Goodman, one of the creators of STaR and a professor at Stanford University, once said that STaR allows the AI ​​model to "guide" itself to a higher level of intelligence by repeatedly creating its own training data. This process is just like humans deepening their cognition. Continuous trial and error and growth can theoretically allow language models to achieve intelligence beyond human beings.

If AI achieves the evolution of logical reasoning capabilities, it means that AI models will be able to do something that has been unable to do so far: conduct in-depth research independently. OpenAI hopes this innovation will significantly improve the reasoning capabilities of its AI models.

Why is reasoning ability so important?

OpenAI CEO Sam Altman said earlier this year that "the most important advances in AI will be around reasoning capabilities."

Reasoning is the key to AI achieving human or superhuman level intelligence. While large language models can already summarize dense text and compose beautiful prose faster than humans, the technology often falls short in solving common-sense problems that seem intuitive to humans, such as identifying logical fallacies and playing tic-tac-toe. chess. When a model encounters this kind of problem, it often "hallucinates" false information.AI researchers interviewed by Reuters generally agreed that in the context of AI, reasoning involves forming a model that enables the AI ​​to plan ahead, reflect how the physical world works, and reliably solve challenging multi-step problems. Improving the reasoning capabilities of AI models is seen as key to unlocking model capabilities, from making major scientific discoveries to planning and building new software applications.

OpenAI’s five-level roadmap for future AI development

OpenAI also announced a five-level roadmap for future AI development on the 11th. According to OpenAI’s conjecture, future AI will go through five stages:

According to the information collected so far, the "Strawberry" project is likely to be the key to helping OpenAI achieve second-level AI. OpenAI believes that Strawberry is approaching level 2, which involves solving problems and is similar to a PhD level without the help of tools. The framework aims to provide a structured approach to understanding and developing artificial intelligence systems that will ultimately surpass human intelligence.

If OpenAI succeeds, the "Strawberry" project is likely to redefine the capabilities of AI, enabling it to make major scientific discoveries, develop new software applications, and perform complex tasks autonomously. Then humans will be one step closer to AGI.