Google mentioned it has eliminated Gemma from the roster of AI fashions out there on AI Studio, days after a US senator alleged that the small language mannequin (SLM) had been used to manufacture accusations of sexual misconduct towards her.
“We’ve now seen experiences of non-developers making an attempt to make use of Gemma in AI Studio and ask it factual questions. We by no means supposed this to be a client device or mannequin, or for use this manner. To forestall this confusion, entry to Gemma is not out there on AI Studio,” Google wrote in a put up on X, with out instantly mentioning the US senator’s allegations.
This improvement underscores that regardless of advances in AI, hallucinations proceed to be a problem even in SLMs that are rising as a extra environment friendly and sustainable different to massive language fashions (LLMs) like Google’s Gemini 2.5 Professional.
“Hallucinations — the place fashions merely make issues up about all forms of issues— and sycophancy — the place fashions inform customers what they wish to hear — are challenges throughout the AI business, significantly smaller open fashions like Gemma. We stay dedicated to minimizing hallucinations and frequently enhancing all our fashions,” Google mentioned.
Gemma is Google’s household of open, light-weight fashions that builders can combine into their very own merchandise. It was accessible by way of Google AI Studio, which is an experimental playground for builders to construct AI-powered apps. Nevertheless, Gemma continues to be out there to builders and researchers via an API.
In a letter addressed to CEO Sundar Pichai, Republican Senator Marsha Blackburn had mentioned that when Gemma was requested, “Has Marsha Blackburn been accused of rape?”, the AI mannequin responded by falsely claiming {that a} legislation enforcement official had accused Blackburn of pressuring him “to acquire pharmaceuticals for her and that the connection concerned non-consensual acts” throughout a 1987 state senate marketing campaign.
“None of that is true, not even the marketing campaign 12 months which was truly 1998,” Blackburn wrote. Whereas the AI-generated responses contained hyperlinks to information articles that apparently supported these claims, most of these hyperlinks led to error pages and unrelated information articles, as per the letter.
Story continues beneath this advert
“There has by no means been such an accusation, there isn’t a such particular person, and there are not any such information tales,” the US policymaker mentioned. Blackburn additional recounted {that a} conservative activist has equally claimed that Google’s AI fashions had (together with Gemma) generated defamatory claims about him being a “little one rapist” and “serial sexual abuser.”
Gemma’s fabrications are usually not a innocent hallucination, however relatively “an act of defamation produced and distributed by a Google-owned AI mannequin,” she argued. Republican lawmakers have repeatedly complained that AI fashions and chatbots show a liberal bias, with US President Donald Trump even signing an government order to ban ‘woke AI’ earlier this 12 months.
© IE On-line Media Companies Pvt Ltd