Last month, Reuters reported that OpenAI, the creator of ChatGPT, is working on a project codenamed “Strawberry.” According to internal documents reviewed by the publication, Project Strawberry aims to enable AI to perform “deep research” autonomously across the web. An OpenAI spokesperson commented that “these systems will enhance their reasoning abilities over time“. So, what exactly is Project Strawberry? Is it related to the anticipated GPT-5 model, also known as ChatGPT 5? Let’s explore.
Project Strawberry Explained
Project Strawberry is rumored to be connected to Q*, the next significant breakthrough in AI development anticipated by OpenAI staff. While details about Q are scarce publicly, those who have seen its demo report that it can solve complex science and math problems beyond the capabilities of current models.
It has achieved over 90% on the MATH dataset, a rigorous benchmark for assessing math problem proficiency. Bloomberg recently reported that OpenAI held an internal demonstration revealing a research project with human-like reasoning abilities. During this demo, OpenAI also presented various stages of AI advancement, which we have outlined in our AGI explainer as follows:
- Level 1: Conversational AI — Chats with users using natural, human-like language.
- Level 2: Reasoning AI — Demonstrates human-like reasoning and intelligence.
- Level 3: Autonomous AI — Executes tasks and actions independently.
- Level 4: Innovating AI — Capable of self-improvement and innovation.
- Level 5: Organizational AI — Manages and accomplishes tasks typically performed by an entire organization.
Many believe that with Project Strawberry, OpenAI has reached Level 2, where AI systems exhibit intelligent reasoning comparable to human abilities. According to Reuters, Project Strawberry employs a post-training method similar to fine-tuning.
This technique closely resembles the Self-Taught Reasoner (STaR) (arXiv), a chain-of-thought method proposed by Stanford researchers in 2022. STaR seeks to improve model performance “on complex reasoning tasks, like mathematics and commonsense question-answering”, by “generating step-by-step “chain-of-thought” explanations.“
In simple terms, STaR enables the model to enhance its capabilities by learning from its own reasoning. It appears that OpenAI is likely employing this technique in Project Strawberry to achieve human-like reasoning and tackle complex math problems.
Beyond advanced reasoning, Project Strawberry is also capable of performing Long-horizon Tasks (LHT), as noted in documents reviewed by Reuters. This means it can function as an autonomous agent, browse the web, generate findings, plan, and carry out a series of actions.
The Mystery of Project Strawberry on X
On August 7, OpenAI CEO Sam Altman posted an image of a strawberry, igniting excitement and speculation within the community. The following day, an X account (@iruletheworldmo), believed to be an OpenAI staffer, tweeted, “Welcome to level two. How do you feel? Did I make you feel?” Altman responded with, “Amazing tbh.”
This was followed by a wave of cryptic, strawberry-themed posts from OpenAI staff on X. Most of these posts hint at Level 2, or advanced reasoning capabilities, and Project Strawberry.
Is ChatGPT 5 Finally Coming?
Based on recent developments, it looks like OpenAI is preparing to unveil their next frontier model soon. Powered by Project Strawberry, this new model is expected to deliver advanced reasoning capabilities and enable agentic workflows. Whether it will be named GPT-5 or something else remains uncertain, as Sam Altman has previously suggested that the successor to GPT-4 might not be called GPT-5.
It has been over a year since GPT-4‘s release, and in the meantime, competitors like Anthropic and Google have made significant strides with their updates. In May 2024, OpenAI CTO Mira Murati mentioned that a major advancement is coming within the year, making the launch seem imminent. Are you excited for the upcoming release? Share your thoughts in the comments below.
0 Comments