Google responds to Sen. Blackburn in AI "hallucination" clash
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Lindsey Bailey/Axios
Hallucinations across large language models are a problem for the entire AI industry and problems arise when consumers use models only meant for developers and researchers, Google wrote in a letter to Sen. Marsha Blackburn (R-Tenn.) obtained by Axios.
Why it matters: Conservative claims of Big Tech censorship are shifting to focus on AI models that spit out false information — and this is likely just the first of many spats.
- Blackburn has been railing against Google after accusing its AI model Gemma of fabricating accusations of sexual misconduct against her, among other falsehoods.
What they're saying: Google's VP of government affairs Karan Bhatia wrote to Blackburn on Thursday: "We want to be clear—the hallucinations you have seen in Gemma are technical issues common to large-language models impacting users across the political spectrum."
- Bhatia wrote: "You can see this by substituting a non-partisan name, or name associated with different political views."
- Problems arise when consumers use "smaller open models like Gemma, which are not designed to be factual assistants, but rather to be smaller tools for developers to customize and tune, and when users enter a misleading premise."
In response, Blackburn said in a statement to Axios:
- "Google's response is a desperate attempt to dodge accountability and blames the fabrication of heinous criminal allegations on 'technical issues.' Google's unacceptable word salad of excuses fails to address how this happened and how the company will prevent it from happening in the future."
- "Google has known about Gemma's harmful hallucinations for years but chose to turn a blind eye until the company needed to save face."
The big picture: AI companies have been struggling to defend their models against allegations of censorship and bias.
- Chatbots aim to have content rules that keep users safe and produce truthful outputs, and allegations of so-called "woke" AI from President Trump and his allies have sometimes complicated that as companies try to please everyone.
The bottom line: Google's explanation hasn't budged Blackburn, who has been rallying against Big Tech for children's safety issues and alleged censorship for years and is unlikely to stop now.
