OpenAI Strawberry Project Introduces New Forms of AI Reasoning 

Open AI has introduced its new OpenAI Strawberry AI model, “o1 advancing reasoning in AI that outperform older models.  

Described as a “preview,” this release also launched a smaller, cheaper version, “o1-mini.” The newly developed model can now solve complex questions and problems that were until now out of the range of earlier models. It will do so faster than a human, pushing AI reasoning further to reach human-like artificial intelligence. 

Solving Problems Through Reinforcement Learning 

According to the AI company, the OpenAI Strawberry model is more capable of solving multi-step problems and writing code than its predecessor, GPT-4o. What sets this new model apart is that it employs AI reinforcement learning, a training technique that rewards or penalizes the AI for its actions to learn even more. 

OpenAI’s chain-of-thought reasoning enables these latest models to sequence the problems step-by-step so that it generates a more accurate and human-like approach to problem-solving.  

In other words, the AI chain of thought enables these models to process tasks step by step and makes them superior to earlier models such as GPT-4. “The model excels in competitive programming and math,” OpenAI says, letting users follow the reasoning behind each answer. 

While OpenAI Strawberry improves problem-solving, especially in domains involving math and coding, it still is slower and more expensive compared to previous models.  

Advanced Reasoning, Yet Hallucinations  

The training method of o1 is completely different, says OpenAI’s research lead Jerry Tworek. It is not like the previous versions, which were trained to just mimic patterns in data, it has been mainly trained on a curated dataset.  

Tworek claims that o1 “hallucinates” less than previous models, but that problem is far from solved, proving that even with the reasoning capabilities it is hard to deal with AI hallucinations. 

In this demo, OpenAI’s chief research officer, Bob McGrew, gave o1 a math puzzle to show its reason, and the model answered this correctly-after 30 seconds-and showed its line of reasoning, very similar to human-like thinking. 

The OpenAI Strawberry project is a further step toward the creation of how artificial intelligence systems think exactly like human beings do. While o1 can neither search on the web nor process images, its reasoning capabilities is paving the way for new opportunities in AI research. 

“We have been spending many months working on reasoning because we think this is actually the critical breakthrough,” McGrew said.  

The focus on reasoning is significant because OpenAI tries to make its AI more intelligent and helpful in medicine, engineering, and other important domains. 

Despite several limitations, OpenAI Strawberry represents a significant advance toward AI that can think more like humans. 

Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Tech sections to stay informed and up-to-date with our daily articles.