Google’s AI will probably be able to be “uncensored” to deliver special content
30 December 09:40
Google may soon provide users with the ability to control Gemini’s responses to sensitive content. This “content filter” will probably allow controlling the level of self-censorship of the model. This is stated in an article by Android Authority, Komersant ukrainskyi reports.
According to the publication, the Gemini API already has settings for “categories of malicious content” available to developers. It is very likely that these settings will soon become available to users. In practice, this will mean that AI will be able to display sensitive content, such as pornography, at the user’s request, similar to the way the search engine does it today.
Google is working hard to improve its AI models and earlier this month announced Gemini 2.0. However, due to the power that AI offers, companies like Google spend a lot of time thinking about what safeguards should be implemented in their systems. A new way in which Gemini could soon give users more control over the type of content it generates is being considered.
Follow us on Telegram: the main news in a nutshell
In version 15.51.24.sa.arm64 of Google’s Android beta app, a new option in Gemini’s settings was discovered that appears to be preparing for the ability to filter Gemini’s output by content. At the moment, selecting this option doesn’t provide any useful result and only redirects to a URL on Google’s Gemini site, which is not yet live.
Google already has some official rules about what can be done with Gemini, and it also has goals about what content Gemini tries to avoid. Specifically, developers have had access to Gemini’s fairly detailed security settings, giving them control over several “categories of harmful content,” including hate speech, harassment, and explicit sexual content.
Knowing all of this, it is possible that Google could extend some level of the same control to Gemini end users on Android and the web version. However, Google could take a number of different paths to implement this feature, from the broad developer-level filters that the API currently has to even more restrictive ones, all with the goal of preventing Gemini from distributing particularly objectionable content.
There is currently no specific information on when this new option might become available.
