Home entertainment After years of closure .. Openai opens the door to free artificial...

After years of closure .. Openai opens the door to free artificial intelligence

4
0

Openai has launched two new sophisticated models of artificial intelligence bearing the name GPT -SS-120B and GPT -SS-20B, in a move aimed at expanding the scope of the use of these technologies freely and available to all, and this announcement comes months after the Chinese “Deep Cick” company climbed to the interface thanks to its open models.

Production of texts and the implementation of complex tasks

The new model allows the creation of smart texts in response to orders, as well as their ability to carry out more complex tasks such as writing codes and searching for information on the Internet on behalf of the user.

The “Huging Face” platform hosts these two models to provide direct and easy access to developers and researchers.

Focus on open weights, not full transparency

Although Openai provides “open weights” for the two models, which are the digital values that are modified within the form during training, they do not reveal the training data itself, which makes these models almost open source and are not completely open as it is assumed by the name, and this is a similar trend to what the company “Mita” followed in Llama models.

After years of closure .. Openai opens the door to free artificial intelligence

Play icon

Return of closure and restore the initiative

This step is from Openai the most important since the launch of the GPT-2 model in 2019, after years of the company relied on closed models and fortifying its data for reasons related to safety and commercial monopoly.

It seems that the success of other open models prompted the company to reconsider its strategy, especially with the increasing American invitations to adopt the open artificial intelligence approach.

Declaration of plans for use in companies and governments

Openai aims to use the two models in the private and public sectors, from major companies such as Orange and Snowflake, to government institutions that want to operate artificial intelligence models locally on their servers.

The model is characterized by high operating efficiency, as the “20B” model can work on a 16 GB of memory, while “120B” works on only one graphic processing unit.

Before the full launch in order to preserve safety

Despite its announcement since July, the company postponed the official launch of the largest model until the completion of additional safety tests, and CEO Sam Altman explained that the release of open -weight models is an irreplaceable decision, stressing the company’s desire to act with responsibility despite the great community enthusiasm.

source

LEAVE A REPLY

Please enter your comment!
Please enter your name here