Due to OpenAI’s mistake, users managed to test the o1 artificial intelligence model

OpenAI company wants to release the full version of o1 artificial intelligence model with reasoning capability this year, but some users managed to test this model with a simple trick and tried its advanced capabilities.

According to Tom’s Guide, last Friday, users were able to access the full version of o1 by changing a parameter in the OpenAI addresses. The new model is able to analyze images and access tools such as web search and data analysis.

Since the release of the o1-preview model, OpenAI insiders have been boasting about the model’s full capabilities after removing the preview tag. It was said that the preview model was trained on the previous version of GPT models, but the full model was trained from scratch.

For example, a user was able to use the full version of o1 to solve a picture puzzle. The AI ​​spent about 2 minutes thinking about the problem, but the images and other multimedia data it was fed showed high performance.

In September, the company OpenAI introduced a new type of artificial intelligence model that needs time to think before responding. This feature was added to ChatGPT in the form of o1-preview and o1-mini, none of which showed the full capabilities of the final o1 model, but they were a great improvement over GPT-4 in terms of accuracy.

Better performance of o1 than other OpenAI artificial intelligence models

Sam Altman, CEO of OpenAI, says o1 differs from GPT-style models. Tom’s Guide media claims that during a briefing with OpenAI, they were told that the full version of o1 is a significant improvement over the preview version, and the current report seems to confirm this.

An OpenAI spokesperson said about user access to the o1 model: “We were preparing limited external access to the OpenAI o1 model and ran into a problem. “Now this problem has been solved.”