OpenAI has launched its latest artificial intelligence (AI) model, which is designed to "think before it answers," introducing a level of analysis previously unseen in AI technologies.
In a blog post dated September 12 , OpenAI announced the introduction of several new AI models under a revamped naming system, starting with the o1 series.
"For complex reasoning tasks, this is a significant advancement and represents a new level of AI capability," the company stated. "Given this, we are resetting the counter back to one and naming this series OpenAI o1."
One of the defining features of the o1 model is its ability to engage in "chain-of-thought" reasoning. This approach allows the model to think through problems step-by-step, which is particularly effective in science, coding, and math.
Data released by the company indicates that o1 surpasses GPT-4o in various benchmarks, including math and code competitions, as well as PhD-level science questions.
OpenAI demonstrated the model's capabilities by releasing videos where the o1 model solves a complicated logical problem, successfully codes a video game, and translates a corrupted sentence. However, its accuracy in non-STEM subjects like historical dates, biographies, and trivia is similar to that of smaller language models, such as the GPT-4o mini.
ChatGPT Plus subscribers now have access to the "preview" and "mini" versions of the o1 model. OpenAI has indicated that more enhanced versions are in development and expected to roll out in the coming months.
The introduction of the o1 model represents a leap forward in AI, potentially revolutionizing problem-solving across multiple domains.
In other news, the GPT-4o model has been assessed as posing a "medium risk" in its potential to influence political opinions through generated text .
