OpenAI launches new super AI with 'reasoning' capabilities.

On September 12th, OpenAI launched a series of new artificial intelligence (AI) models capable of spending more time thinking in order to provide more accurate and beneficial responses to users.

The new models, known as OpenAI o1-Preview, are designed to solve more complex tasks and challenging problems in science , coding, and mathematics—fields where previous models have often been criticized for not providing consistent answers.

OpenAI's new AI model is capable of 'reasoning' and can solve even Olympic-level math problems. Photo: AFP

OpenAI o1-Preview was trained to fine-tune its thought process, experiment with different methods, and detect errors before providing a final answer.

Sam Altman, CEO of OpenAI, assessed the new models as having “complex reasoning capabilities,” although they may still have many shortcomings and limitations.

Accordingly, OpenAI is still working to improve the capabilities of its AI models to overcome the "hallucination" problem – a phenomenon where chatbots create persuasive but inaccurate content.

OpenAI researcher Jerry Tworek said the new model will cause fewer "illusion" issues, but it doesn't completely solve the problem.

Reportedly, OpenAI o1-Preview models have successfully solved challenging requirements in the fields of physics, chemistry, and biology with a level of expertise equivalent to that of PhDs.

In particular, in the fields of mathematics and coding, OpenAI o1-Preview achieved an accuracy rate of up to 83% when solving problems in the International Mathematical Olympiad. This figure is significantly higher than the 13% accuracy rate of the previous GPT-4o model.

Sam Altman, CEO of OpenAI, assessed the new models as having "the ability to perform complex reasoning." Photo: AFP

According to OpenAI, this new reasoning capability could assist healthcare researchers in annotating cell sequencing data. Simultaneously, the tool could support physicists in developing complex formulas.

OpenAI also stated that the new AI models have passed rigorous jailbreak tests and can resist attempts to bypass security barriers.

In addition, safety measures have been strengthened, including recent agreements with the US and UK AI Safety Institutes, which have been granted early access to these models for testing and evaluation.

Source: https://laodong.vn/cong-nghe/openai-ra-mat-sieu-ai-moi-voi-kha-nang-lap-luan-1393825.ldo