News in English

OpenAI Preps AI Model With Improved Humanlike Reasoning

OpenAI is reportedly readying an artificial intelligence (AI) model offering more evolved humanlike reasoning.

The company’s o3 model devotes more time to computing an answer before replying to user queries, with the aim of resolving more complicated, multistep problems, Bloomberg News reported Friday (Dec. 20) from an OpenAI livestream event.

During its event, OpenAI offered some details on how o3 is superior to o1, the reasoning model launched in September. The company also wants safety and security researchers to apply to test the models before unveiling new software, the report added.

OpenAI plans to release a smaller version of the model, dubbed o3-mini, at the end of January and the o3 model soon after that, CEO Sam Altman said during the event. A company spokesperson says there is no o2 model, out of respect” for the British telecommunications firm of the same name.

As Bloomberg noted, the launch comes as other leading AI companies are moving forward with increasingly advanced versions of their models. 

For example, Google recently rolled out a new version of its flagship model, Gemini, said to be twice as fast as the past model and can “think, remember, plan and even take action on your behalf.” And Meta recently hinted at plans to put out its Llama 4 model in 2025.

Still, the report added, a number of these companies are seeing diminishing returns from their costly efforts to create newer models. Tactics such as putting  greater focus on “reasoning” help them get around the challenge of finding new, untapped sources of high-quality, humanmade data for training, Bloomberg said.

In other AI news, PYMNTS wrote last week about an AI system developed at MIT that could let warehouse robots deftly handle odd-shaped packages and navigate crowded environments without putting human workers in danger. 

Although robots excel at repetitive tasks such as moving pallets, the new PRoC3S technology could finally solve the long-standing problem of robots safely performing more complex warehouse jobs requiring human dexterity and spatial awareness.

“In theory, PRoC3S could reduce a robot’s error rate by vetting its initial LLM [large language model]-based assumptions against more specific and accurate understandings of the warehouse environment,” Erik Nieves, CEO and co-founder at Plus One Robotics, told PYMNTS.

“Think about it like this: A warehouse robot operating solely on LLM guidance has been described how to complete a task. The PRoC3S concept goes one step further by placing a digital robot in a simulated environment of that task. It’s essentially the difference between classroom instruction and a really good field trip.”

The post OpenAI Preps AI Model With Improved Humanlike Reasoning appeared first on PYMNTS.com.

Читайте на 123ru.net