OpenAI has released several new artificial intelligence models under a revised naming scheme — starting with its latest OpenAI o1 model it says can “think before it answers.”
“For complex reasoning tasks, this is a significant advancement and represents a new level of AI capability,” OpenAI said in a Sept. 12 blog post. “Given this, we are resetting the counter back to one and naming this series OpenAI o1.”
The new models can take their time to think and use “chain-of-thought” reasoning to solve complex tasks — particularly in STEM (science, technology, engineering and math) and coding-related tasks, OpenAI said.
Source: OpenAI
The AI firm shared videos of OpenAI o1 coding a video game from a prompt and solving a complex logical puzzle, among other things.
The OpenAI o1 “preview” and “mini” models were made available to ChatGPT Plus subscribers with the firm planning to release improved versions in the coming months.
OpenAI shared data suggesting OpenAI o1 defeats GPT-4o in several benchmarks, including PhD-level science topics in Biology, Chemistry and Physics and some United States high school exams.
OpenAI o1 improvement model compared with GPT-4o on several benchmarks. Source: OpenAI
Due to OpenAI o1 mini’s focus on STEM reasoning capabilities, it isn’t as knowledgeable in other areas outside of its narrow focus, OpenAI said.
“[Its] factual knowledge on non-STEM topics such as dates, biographies, and trivia is comparable to small LLMs such as GPT-4o mini.”
“We will improve these limitations in future versions, as well as experiment with extending the model to other modalities and specialties outside of STEM,” it added.
Industry pundits anticipated OpenAI would release a reasoning-focused AI model in September under the codename Strawberry.
However, OpenAI doesn’t disclose distinctions between different models under development.
Magazine: $1M bet ChatGPT won’t lead to AGI, Apple’s intelligent AI use, AI millionaires surge: AI Eye