Openai finally reaches its name level, and it will drop new open source models

For the first time in five years, Openai has released new, lightweight, lightweight models designed to merge them easily into other programs.

in Blog post On Tuesday, the GPT -SS-120B and GPT -SS-20B described it as flexible but strong algorithms that can perform a variety of tasks and use them in many settings. The company also included the reaction portal and more Wide blog This explains the models and how they work. Openai CEO, Sam Altman, He said via x On Tuesday, artificial intelligence would help “new types of research and create new types of products.” It seemed that Al -Taman highlights the open source method: “We believe in individual empowerment.” “Although we believe that most people want to use a comfortable service like Chatgpt, people should be able to directly control and modify artificial intelligence when they need it, and the advantages of privacy are clear.”

Unlike most Chatgpt products, open source models reveal the training parameters used to build their system. This level of full -time transparency gives benefit from knowing how the system works and why it might act in the ways it does. The last time Openai has released an open model for the source during the first Trump presidency (really, it seems to be decades ago), with GPT-2 launch. It was when Openai was still an unknown start, and it would take several years ago Chatgpt launch In 2022, again in those days, the company was still routinely referred to by a title like “like”Elon Musk project of artificial intelligence,Despite the fact that Musk has been separated from ways with the company.

Perhaps more importantly, you will not have to pay ten cents to use these models. And as long as your computer specifications have ended, you can operate locally instead of relying on Openai servers. Any of the emergence of the Day of Resurrection in AI may want to look at it.

From the appearance of things, there are a lot of promising things in the new versions of the company. Openai says, GPT-SS is designed to integrate into the workflow of agents, which means that New types of automated work-I am Through the so -called “agents”It can be supported by new algorithms. The new models are also included under the APache 2.0 license, which allows users to create a new program with algorithms without worrying about prosecution. “My son is free without worrying about LifeteTle’s restrictions or patent risks – whether you are experimenting, allocating or publishing commercially,” he writes Openai. The ecological system is largely open source on a large number of these licensing agreements, allowing companies to build free models.

Openai also pushed some lip services for artificial intelligence safety in its announcement. The company claims that, in addition to “playing models through comprehensive training and evaluation of safety”, it also provided “an additional layer of assessment Reading frameworkIt is designed to evaluate and track risk -framed behavior in large language models.

In recent years, a lot of criticism has been aimed at Openai about its decision to continue the “walled garden” approach to software development. The company’s LLM versions remained royal, and therefore, they were stopped from the general inspection. Now, Openai is clearly trying to prove that haters are wrong and are bound by his commitment to being a “really open” organization. It is guessing anyone about whether the company will be able to maintain such a commitment to FOSS ethics, given that it is also organized It deserves hundreds of billions of dollars On paper. The money organized behind Openai may continue to see interest in having exclusive and closed systems, and most importantly, it is only controlled by a selection of the informed.

It is worth noting that the GPT-5, the strongest new and very expected model, will be almost released in the same way as the last GPT version: closed.