OpenAI officially announces the update of GPT4o! Is the legendary "Strawberry" here?

OpenAI is up to something again.

On the 13th, ChatGPT posted on social media X, stating that the GPT-4o model of the chatbot family has been updated again, emphasizing that this time it will be very different. This model is not the latest API-end GPT-4o 0806 model from OpenAI. Currently, even free users can use this model, but with a limit on the number of uses.

Before the official announcement, there were rumors from foreign media that OpenAI will soon announce the "Strawberry" project in the ChatGPT update.

After the official announcement, netizens are also speculating that this new GPT-4o model might be "Strawberry".

These speculations are not groundless.

ChatGPT had previously posted a strawberry emoji on X, seemingly hinting at something.

And OpenAI's CEO Ultraman had also shared a picture of strawberries in a summer garden on social media, sparking people's imagination.

So what exactly is this "Strawberry" project?

Last month, OpenAI's internal team was revealed to be developing the "Strawberry" project, aiming to enhance OpenAI's model's reasoning ability, ability to handle complex scientific and mathematical problems, enabling large models to not only generate query answers but also plan ahead for autonomous, reliable internet browsing, and conduct "deep research" as defined by OpenAI.

The predecessor of this "Strawberry" project is the Q* algorithm model, which can solve challenging scientific and mathematical problems. Mathematics is the foundation of generative AI development. If AI models master mathematical abilities, they will have stronger reasoning capabilities, even comparable to human intelligence. This is something that current large language models are unable to achieve Ultraman once emphasized that the key to the future development of AI will revolve around reasoning ability.

According to informed sources, the "Strawberry" project includes a specialized "post-training" method, where generative AI models have undergone pre-training on a large dataset and further adjusted to improve their performance on specific tasks. This is similar to the "Self-Taught Reasoner" (STaR) method developed by Stanford University in 2022.

One of the creators of STaR, Stanford University professor Noah Goodman, has stated that STaR allows AI models to create their own training data repeatedly, "guiding" themselves to higher levels of intelligence, theoretically enabling language models to surpass human intelligence.

This aligns with OpenAI's pursuit of "reasoning ability". Furthermore, on the 11th, OpenAI also unveiled a five-level roadmap for the future development of AI:

According to OpenAI's speculation, future AI will go through 5 stages:

Level 1: Chatbots, artificial intelligence with conversational language

Level 2: Reasoners, human-level problem solvers

Level 3: Agents, systems capable of taking action

Level 4: Innovators, AI that can assist in invention

Level 5: Organizers, AI capable of performing organizational tasks

Based on the information currently gathered, the "Strawberry" project is highly likely to be crucial in helping OpenAI achieve Level 2 AI.

If OpenAI succeeds, the "Strawberry" project is likely to redefine the capabilities of AI, enabling it to make significant scientific discoveries, develop new software applications, and autonomously perform complex tasks, bringing humanity closer to AGI once again