Google CEO Sundar Pichai supported NVIDIA’s joint letter defending vulnerability-heavy AI models. Pichai stated that Google has been benefiting from the open source ecosystem for many years and constantly contributes to this field. Also developed by Google DeepMind Gemma models with open weights He reminded that he published it regularly. 24 July The dated letter considers open models a key element for U.S. AI leadership.
The letter published by NVIDIA CEO Jensen Huang is titled “Open Weights and American AI Leadership.” The text compares today’s choice to the open source movement that transformed the software world in the 1980s. Accordingly, developers can download, examine, modify and run vulnerability-weighted models on their own infrastructure. Thus, organizations can use options that suit their needs without depending on a large model provider for each transaction.
The letter argues that open models will leave AI out of the hands of just a few large companies. Therefore, startups, universities, public institutions and small businesses can more easily access advanced models. Instead of training models from scratch, organizations can adapt ready-made weights according to their own data and tasks. In addition, factories, hospitals, farms, classrooms and local businesses can develop solutions suitable for their own infrastructure.
Open weights also increase companies’ technical control over models, according to the statement in the letter. Full dependence on closed services can tie customers to a single provider and move internal knowledge outside the organization. On the other hand, open models allow companies to store the customizations they develop in their own systems. Signatories state that this option will increase competition between cloud services, processors, applications and artificial intelligence services.
Vulnerability models expand competition and security
joint letter, open weighted artificial intelligence models He argues that it reduces costs and accelerates innovation. By choosing small models appropriate to the task, developers can tackle certain tasks with fewer computing resources. In fact, the text also discusses the dependence of institutions on a single closed service in terms of technological sovereignty. Signatories want organizations to own their models, improve them and retain the talent they have accumulated.
The security section directly acknowledges the risks that open models carry. Once a company publishes the weights, it cannot easily track modified copies or retract entire samples. However, the letter argues that these risks do not justify broad restrictions. He states that while cyber attackers use advanced artificial intelligence tools, defense teams also need to have access to similar capabilities.
Signatories say transparency gives researchers the opportunity to find and fix vulnerabilities faster. Large developer communities can contribute to the development of protection methods by testing models under different conditions. However, the letter challenges the assumption that closed models are safer per se. The text defines the audit carried out by many researchers as one of the important parts of security studies.
Pichai’s statement re-emphasized Google’s open weight training with the Gemma family. Google DeepMind makes the weights of Gemma models prepared by using Gemini research available to developers. The company aims to make the models work locally on phones and computers by preparing versions suitable for different hardware classes. Thus, developers can run certain AI tasks on the device without sending data to remote servers.
Introduced by Google in April Gemma 4 familyIt contains options with 31 billion, 26 billion and 12 billion parameters. The company is positioning the 12 billion-parameter version for multi-modal operations and native applications on laptops. Additionally, quantization-focused training checkpoints reduce the amount of memory required by models. This technique provides more efficient local use on phones and computers with appropriate hardware.
In addition to NVIDIA, Meta, Microsoft, Mistral AI, Hugging Face, IBM and Mozilla also signed the letter. Dell Technologies, Andreessen Horowitz, Perplexity, ServiceNow and Y Combinator are also among the organizations that supported the joint call. Although the name OpenAI appears in the text dump of the PDF, this company’s logo is not included in the visual signatory list. Therefore, the list with the logo on the last page of the letter is used as the main source for verifying the signatories.
The joint call asks policymakers not to impose premature and broad restrictions on open models. He also wants them to pave the way for investments that will increase access to computing resources for startups and researchers. The text calls for greater support for common datasets, assessment tools, and application development resources. Pichai shared his X post with the caption “Onwards!” He concludes by saying that Google will continue its work on Gemma.
Join Channel