“The vast majority of our [enterprise and startup] Customers already use many open models, “said Casey Dvorak, a research program manager at OpenAI, in a media briefing about the model release.” Because there is none [competitive] Open model from OpenAi, we wanted to connect that gap and actually allow them to use our technology across the board. “
The new models come in two different sizes, the smaller of which can be theoretically performed on 16 GB RAM – the minimum amount that Apple is currently offering on its computers. The larger model requires a high-end laptop or specialized hardware.
Open models have a few important usage scenarios. Some organizations may want to adjust models for their own purposes or save money by performing models on their own equipment, although that equipment has considerable costs in advance. Others – such as hospitals, law firms and governments – need models that they can perform locally for data security reasons.
OpenAI has facilitated such an activity by releasing its open models under a permissive Apache 2.0 license, with which the models for commercial purposes can be used. Nathan Lambert, lead after training at the Allen Institute for AI, says that this choice is commendable: such licenses are typical of Chinese open model releases, but Meta has released its LAMA models under a customized, limiting license. “It’s very good for the open community,” he says.
Researchers who study how LLMS works also need open models so that they can investigate and manipulate those models in detail. “This is partly about re -changing the Dominance of OpenAI in the research ecosystem,” says Peter Henderson, a university teacher at Princeton University who has worked extensively with open models. If researchers take over GPT-Oss as new work horses, OpenAI could see a number of concrete benefits, says Henderson-It could take over innovations discovered by other researchers in its own model ecosystem.
More in general, says Lambert, can now release an open model to restore its status in an ever -drank AI environment. “It goes back a bit to years ago, where they were seen as the Ai Company, “he says. Users who want to use open models now have the option to meet all their needs with OpenAI products, instead of turning to Lama from Meta or Qwen from Alibaba when they have to perform something locally.
The rise of Chinese open models such as QWen in the past year may have been a particularly striking factor in the Calculus of OpenAi. During the media briefing, an employee of OpenAi emphasized that the company does not see these open models as a response to actions undertaken by another AI company, but OpenAi is clearly tailored to the geopolitical implications of China’s open model. “Broad access to these capable open-weight models made in the US help to expand democratic AI rails,” the company wrote in a Blog post announcing the release of the models.
#OpenAi #finally #released #openweight #language #models