When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
Artificial intelligence (AI) development has made great strides in a relatively short period of time, but there is still much to be done. Leading products like ChatGPT and Claude Sonnet offer impressive capabilities but with significant logical limitations.
However, this is starting to change. Last week, Reuters reported that OpenAI, the creator of ChatGPT, is working on a new approach to AI models as part of ‘Project Strawberry‘, also known as ‘Q*,’ which will grant large language models (LLMs) improved reasoning.
Although the source familiar with the matter didn’t confirm when Strawberry would be released, they did share internal documentation providing some basic details about the project.
Techopedia followed up on the report for clarification from OpenAI, but the world’s leading AI startup did not immediately respond to comment.
Key Takeaways
- Strawberry is an internal project at OpenAI, according to Reuters).
The model was formerly known as Q*. - OpenAI’s model will reportedly feature enhanced reasoning capabilities — able to handle more complex multi-step tasks.
- This solution could operate more autonomously than existing LLMs.
What We Know About OpenAI’s Strawberry
At the moment, information about Strawberry is scarce, but one of the documents reviewed by Reuters describes a project that uses the Strawberry models to navigate the internet autonomously to perform “deep research.”
This would enable a model to scour the internet, extracting information from articles and other content, which it could then use to continuously improve its reasoning over time.
It is worth noting that one of Reuters sources noted that Strawberry shares similarities with the Self-Taught Reasoner (STaR) technique developed at Stanford in 2022. Under STaR, a model can iteratively create its own training data and become more intelligent over time.
The source also alleged that internal documents suggest OpenAI is designing Strawberry to perform long-horizon tasks — more complex tasks which require a model to perform a series of actions over time. Giving the model the capability to perform such tasks would grant it greater independence than today’s LLMs.
Alon Yamin, co-founder and CEO of Copyleaks, told Techopedia:
“OpenAI’s ‘Strawberry’ project signals a significant stride in AI capabilities, potentially revolutionizing how we interact with generative AI technology and how it solves complex problems. The implications for research, software development, and even scientific discovery are immense.
“Nevertheless, as we embrace this frontier, we must continue to prioritize the implementation of comprehensive guardrails. These guardrails will ensure that AI advancements like ‘Strawberry’ are harnessed responsibly, mitigating potential risks and maximizing their positive impact on society.”
Why Does Strawberry Matter?
If the information reported is correct, then it indicates that OpenAI’s LLM development is evolving so that its AI models will be able to learn more independently, and conduct complex multi-step tasks.
Models using this approach will be able to automate a wider range of tasks than current LLMs, which require a high level of human oversight. For instance, users not only need to enter prompts to use ChatGPT, they also need to fact-check its outputs, to make sure there are no hallucinations and incorrect statements.
This is because LLMS don’t think autonomously like human beings. They’re trained to learn patterns in human language and predict responses to basic inputs. As a result, they don’t think in the way a human would, and they don’t have a notion of common sense or logic.
Let’s Not Get Too Excited Just Yet
Undoubtedly Strawberry will add some innovations to the market, but at this stage it’s important not to get too wrapped up in the hype. After all, ahead of the release of GPT-4o, there were lots of rumors of a release of GPT-5.
While GPT-4o has been a solid multimodal entry to the LLM market, it hasn’t really been able to pull away from the crowd of other popular language models like Claude 3 Opus/Sonnet or Gemini in terms of its performance.
That being said, if Strawberry unveils new techniques that can enhance the reasoning capabilities of LLMS, then it will be a welcome addition to the market, much in the same way retrieval augmented generation (RAG) and other techniques have been.
The Long Term Implications of Strawberry
Until we receive confirmation from OpenAI, one way or the other, it is difficult to identify the long-term implications of Strawberry. Is the project legitimate? Could it be shelved?
But if the information is credible, it would indicate that LLMs are evolving beyond being human-assisted to a more autonomous role, with the ability to create their own training data sets and perform tasks with minimal assistance.
From a top down perspective, the ability to automate the creation of training data could lessen the workload of AI and machine learning researchers, who typically need to curate data sets, while the ability to handle complex automations will open the door to a wider range of use cases in areas like software development.
Such an approach could also increase risk. If models are more independent, there will inevitably be less human oversight and control. This raises questions about whether AI-generated training data and autonomous actions can be kept in line with responsible AI development.
At the very least, Strawberry would be able to answer more complex questions, and would be a small step on OpenAI’s journey to develop artificial general intelligence (AGI), a type of AI that would stand perform in line with human intelligence.
The Bottom Line
The generative AI rollercoaster still has plenty of surprises in store. For now, we’re in a period of calm, but it’s not going to stay that way.
New capabilities will evolve, and the tasks you’re going to be able to automate with AI will inevitably increase.
