OpenAI unveils AI model o1: human-like reasoning and improved logic
OpenAI has unveiled its latest model, o1, designed to mimic human reasoning and provide more logical answers to complex questions. The company aims to achieve this by training the model to integrate previous requests into its reasoning process in rea time. This capability is available to subscribers of ChatGPT Plus and ChatGPT Team, with plans to eventually extend it to the free version of ChatGPT.
To develop o1, OpenAI used a new optimisation algorithm and a specialised training dataset. The model uses reinforcement learning to solve problems independently and then uses a thought chain to process queries in a human-like manner. This approach has potential applications in video game development, as evidenced by OpenAI’s posts on social media.
Initial assessments of o1 indicate promising results with fewer hallucinations compared to previous models. Hallucinations refer to cases where the AI generates responses based on multiple incorrect results without verifying their authenticity. This problem was a criticism of OpenAI’s previous search engine, SearchGPT, which is still undergoing limited testing.
There is a cost associated with the availability of o1 for developers who want to use the API. The cost has increased from GPT-4o, charging $15 per million tokens for searches and $60 per million tokens for replies.
Business AM
Subscribers 0
Fans 0
Followers 0
Followers