OpenAI's "Open AI Model," originally scheduled to be released this week, was earlier announced by CEO Sam Altman.Personally announce the postponement, and explained that the original plan was to release a language model with open weights this week, but considering that the related security risks still need further testing, it was decided to postpone the launch schedule.
Sam Altman said: "Once you release the model weights, there is no going back. We take this experiment very seriously. This is the first time OpenAI plans to release a language model with complete weights, and we want to do it well."
This "open model" is seen as OpenAI's first return to a relatively open technology development strategy since the release of GPT-2019 in 2. The model is expected to be run on users' own hardware, capable of long-term reasoning and highly reliable answers, symbolizing OpenAI's continued exploration of the possibilities of open source and transparency in AI.
However, this project has also sparked considerable discussion and careful evaluation amidst the rapid development of AI technology. Releasing the weights, which exposes the model's internal operating details, could lead to increased risks to society and information security if misused or abused. OpenAI has repeatedly stated that AI safety and application oversight are crucial components of its development strategy.
Although Sam Altman did not specify the specific postponement schedule or disclose the technical bottlenecks encountered in the current testing, he emphasized that the model will be officially released after ensuring that it complies with open security principles.
According to previous public statements, this open weight model is expected to have the following characteristics:
• Applications can be deployed on consumer-grade hardware (such as GPU servers)
• Supports multiple rounds of long-text reasoning and question-answering
• Maintain the logical consistency and high accuracy of OpenAI’s existing language models
• Open user community oriented, facilitating research and model tuning applications
Furthermore, some industry observers suggest that OpenAI's decision to release this model is partly a response to competitive pressure from the current open AI trend. Currently, companies such as Meta, Mistral, and Cohere have successively open-sourced their large-scale language models, and open model communities (such as LLaMA and Mixtral) are rapidly growing, promoting technical exchange and application flexibility among companies, research institutions, and developers.
At present, OpenAI has not announced the specific model name and parameter scale, but it is generally believed that the model may be comparable to Meta's LLaMA 3 series or the latest version of the Mistral model, becoming an important part of the AI open source ecosystem.
As language models continue to advance toward scalability, security, and controllability, OpenAI's postponement, while somewhat conservative, demonstrates the continued importance of risk prevention and long-term thinking in the rapidly evolving AI industry.
OpenAI's acquisition of WindSurf rumored to be broken
On the other hand, in May this year, it was reported that OpenAI planned to acquire AI startup WindSurf for up to US$5 billion to strengthen its AI-assisted coding technology layout. However, the latest news shows that the plan may have changed. The main reason is that two core executives of WindSurf, including WindSurf CEO Varun Mohan and co-founder Douglas Chen, have jumped to Google DeepMind and signed a non-exclusive technology licensing agreement, casting a shadow on this blockbuster acquisition plan.
WindSurf is a startup company that focuses on AI code generation and development tools. It has gained considerable popularity in the developer community over the past year.Bloomberg NewsIt was previously reported that OpenAI and WindSurf had reached a preliminary acquisition agreement, signing a letter of intent and entering exclusive negotiations. The transaction value was expected to be as high as $30 billion. However, concerns within WindSurf regarding the partnership between Microsoft and OpenAI ultimately led to a rift in the negotiations.
According to subsequent reports, Google did not directly acquire WindSurf, nor did it participate in equity investment, but through non-exclusive technology licensing and poaching, it successfully acquired WindSurf's core talent and some technology rights, spending a total of approximately US$24 billion.TechCrunch websiteIt said that the addition of Mohan and Chen will help DeepMind improve its automated development and deployment capabilities of AI agent programs.
