OpenAI seeks to make its upcoming ‘open’ AI model best-in-class

In the end of March, Opena Declaration The purpose of publishing its first “exposed” language model after that GPT – 2 Some time this year. Now the details about that model are starting to get out of the organization sessions with the AI ​​developer community.

Sources tell TechCrunch that OpenAI’s VP Idan Clark is leading the development of the Open model, which is very early. OpenAI aims to release the release in early summer and aims to create the model-an argument on the line of Opina’s O-series models-benchmark-toping between the openly open rational models.

According to the source of the TechCrunch, Open is exploring a highly permissible license for the model with a few uses or commercial restrictions. Have been open models like Lama and Google Jemma Criticized To impose strict requirements – Openly criticism is seemingly trying to avoid.

Opena is facing increasing pressure from rivals like the Chinese AI Lab DIPSEC that has adopted an open approach to launch models. Unlike the OPENY TIGH, these “open” competitors make their models available to the AI ​​community for examination and in some cases commercialization.

It has proved to be a wild successful strategy to some dresses. Open AI Models Invested Lamma Family in Lama Family said that Meta said in the early March that Lama had More than 1 billion downloadsThe Meanwhile, DIPSEC has quickly collected a large world user base and Has attracted the attention of domestic investorsThe

Sources tell TechCrunch that OpenAI wishes for its open model, which will be “text out” to run high-end customer hardware, and perhaps the developers allow the developer to turn on and stop “logic”, like rational models Recently published by anthropologists And others. (Argument can improve accuracy, but at the expense of increased delay)) If the launch is well taken, openAI can follow it with additional models-including small models.

Between Public commentOpenAI chief executive officer Sam Altman says he thinks Open is from the wrong side of history when it is about to open his technologies to sourcing.

“[I personally think we need to] Find out a different open source technique, “Ultman said during a Reddit Q&A in January. [going forward]But we will maintain less than the less leadership than the previous years. “

Ultman also says that the uppaper Open model will be thoroughly reddish and evaluated to protect the Open model. Sources tell TechCrunch that the company wants to publish a model card for the model – a complete technical report shows the results of the internal and external benchmarking and security test of OpenAE.

“[B]The issue of publishing, we will evaluate according to this model [to] Our preparation structure, such as we want for any other model, “Ultman Says in a post at X last monthThe “[A]ND we know that we know that this model will be corrected after re -publishing. “

Some AI ethics for OpenAI raised fuel for ethics Report Protection test of recent models and Failed to reveal model cards for othersThe Ultman too The accused stands OpenAI executives are confusing about model security reviews before that Short Av to November 2023.

We reached OpenAI for comments and if we listen back to this piece will update.