OpenAI reportedly getting ready to launch new ‘Strawberry’ AI mannequin

OpenAI is reportedly getting ready to launch a brand new synthetic intelligence product with superior options that may clear up issues and duties which might be past present AI fashions.

First reported by The Data, the brand new mannequin is internally known as “Strawberry” and is alleged, amongst different issues, to have the ability to clear up math issues it has by no means encountered, carry out high-level duties similar to creating advertising methods, and even clear up complicated phrase puzzles. In a single instance, the brand new mannequin was capable of clear up the New York Instances phrase puzzle “Connections.”

Earlier claims concerning the mannequin embrace that it scores over 90% of the MATH benchmark, a group of championship-level math issues. By comparability, GPT-4 scored solely 53% on the check and GPT-4o achieved 76.6%. As of July, GPT-4o had the best MATH benchmark rating of an AI mannequin at present accessible, which means that if Strawberry delivers as promised, it is going to place OpenAI effectively forward of its opponents.

The place the information maybe turns into extra fascinating is the historical past of its improvement, as a result of as a lot because the title “Strawberry” sounds type of good and noncontroversial, it wasn’t all the time recognized by the title. The mannequin was beforehand recognized internally as Q* (pronounced Q-Star) and it was essential to the transient interval of chaos that descended on OpenAI final yr, together with the ouster of Chief Government Officer Sam Altman, earlier than he returned to the corporate days later.

As reported in November, Altman’s ouster was stated to have been influenced by considerations a couple of main AI breakthrough achieved by the corporate. A gaggle of OpenAI researchers wrote a letter to the board earlier than Altman’s ouster highlighting the potential dangers that might be posed by superior AI and, particularly, Q*, the mannequin now generally known as Strawberry.

Among the many claims made on the time embrace that Q* may characterize a serious breakthrough within the journey towards constructing synthetic common intelligence. AGI is a sophisticated type of AI that possesses the flexibility to grasp, study and apply data throughout a variety of duties, much like human cognitive talents.

The potential danger introduced by AGI comes all the way down to the potential for lack of management over AI programs, resulting in unintended penalties and the potential for AGI creating targets misaligned with human values, which may lead to vital hurt. Put much less technically, there’s a fear that an AGI mannequin may flip into Skynet from the “Terminator” motion pictures.

In keeping with The Data, Strawberry might be launched someday within the fall.

Picture: SiliconANGLE/Ideogram