More than two years after Google has been caught by release by Openai Chatggot release, the company collected pace.
By the end of March, Google has launched Modeling Modeling Ai, Gemini 2.5 Pro, so Leading Industry In many standard measuring standards and mathematical abilities. The launch of only 3 months after technological techniques opened another model, flash Gemini 2.0, that is the condition of the time.
The director of the google and the products for Gemini, tell the technology on the opening of the company is a part of the AI Industry.
Doshi said there is a way to take the right style to make these models to go out, what is right to get to the feedback. “
But the release of the release time lifting time appears to have costs. Coffee Have not published safety reports for its latest formIncluded Gemini 2.5 Pro and Gemini 2.0 flash, raising that the company is highly speed.
Today, it is an important standard for Bontier Ai Labs – including Openai, Openai, and Meta, to report on the safety test. These reports are called “system cards” or “previously identical cards by researchers in the first one to introduce the model 2019 Research Papers, Calling them “methods for responsibilities, transparent, transparent, transparent, and practice in charge of machine learning.”
Doshi told Techcrunch that the company did not publish a model card for Gemini 2.5 Pro because it considers “experimental”. The goal of these experiments is an AI style in a limited mode, gain feedback, and against the model before launching the production, she said.
Google is intending to publish Gemini 2.5 Pro card, depending on Doshi, adding a security test and a team with security.
To Track Google Gemini 2.0 flash, which is generally, lacking the card. Model Modio Modio card Google releases is for Gemini 1.5 ProWhich came out more than a year ago.
System card and models provide helpful – and useless, sometimes – information that the company does not extensively advertising on AI. For example, Openai Card Openai System Openai has released for the reasonable model of O1, revealed the form of companies.
By and big, AI community recognizes that these reports are a great effort to support the security and reporting of the past year. As Brush The previous, Google has told the US government in 2023 that it will publish security report for all major “forms.” Company Make similar commitments To Other governmentPromise “Give public transparency.”
There is an effort to control the federal and state level in the United States to create security reports for AI developers. However, they met with a limited adoption and success. One more notable attempt is the success of Cetoded California Bill SB 1047, which corresponds to the tech industry. Members of the council has also legalized to allow US security institutions, the standard settings of the United States, to establish tips for release mode. However, security institution is now facing possible cuts under Trump Administration.
From all shows, Google is falling in some of its promise to report on the model test at the same. It’s a bad thing, many Expert Arguments – especially while these model will be capable and more complicated.