Google Gemini’s frenzied AI releases seem to sideline AI safety reports

Google Gemini’s frenzied AI releases seem to sideline AI safety reports


Summary

  • Google’s Gemini AI assistant is rapidly evolving to compete with rivals.
  • Safety reports for Google’s AI models are delayed, causing concern.
  • Google promises to release safety reports for its AI models once they are public and stable.

Google has been developing its Gemini AI assistant at a frenzied pace to stand a fighting chance in its rivalry with OpenAI’s ChatGPT and Meta AI. We see new features released every week, and earlier this week, the company rolled out real-time Gemini Live video features and v2.5 of the Pro model for free-tier users. However, in the midst of this frenzied release schedule, Google isn’t keeping up with the safety reports for its AI, creating a cause for concern.

Link Image

Related

Google Gemini collects far more personal data than its rivals, surprising nobody

Its deep integration comes at a cost

As a forerunner of AI apps on Android, Google is under considerable pressure from both users and the industry to deliver AI that actively helps users simplify their chores. In an interview with TechCrunch, the big brand’s director and Gemini’s head of product, Tulsee Doshi, agreed the release cadence is a part of keeping the rivalry alive. However, most users may be uncomfortable if the price they pay is safety and security, and unfortunately, that could be happening.

Google is reportedly running late on its safety reports for models as old as Gemini 2.0 Flash, which launched at the beginning of this year. Interestingly, Google is the one that suggested rating the safety and other parameters of AI models in 2019. In a research paper calling for accountability, the company researchers suggested AI models should be accompanied by model cards containing performance metrics, safety tests, and potential use cases. While these aren’t mandated by industry standards or regulations yet, they are released in good faith right alongside new AI models.

Why is Google running behind?

Safety remains a top priority

An image highlighting Gemini 2.5 Pro's performance against other top models.

Source: Google

Google reportedly said the latest Gemini 2.5 Pro model doesn’t have model cards because the company still classifies it as an Experimental model. However, that isn’t the case with the 2.0 Flash model, and yet, its model cards are awaited. However, Google has previously committed to agencies and governments that it will provide safety reports for publicly available AI releases.

The company reassured TechCrunch that Gemini 2.5 Pro’s cards will go live once the model is public and stable, since safety and adversarial testing is complete already. Gemini 1.5 Pro is the last model with cards available, and the company is still not stating timelines for the release of this documentation. Although most users are absorbed in discovering what the new tech has to offer, Google isn’t shy about handling copious amounts of user data through Gemini, and slacking on safety reports isn’t a good look for concerned users.



Originally Appeared Here