Google tells contractors to rate Gemini AI responses outside expertise, TC says

Behind the development of generative AI systems are “prompt engineers” and analysts, who rate the accuracy of chatbots’ outputs to improve the AI system, Charles Rollet of TechCrunch reports. Google currently has contractors working with GlobalLogic to improve its Gemini system and these contractors were originally able to “skip” certain prompts if the AI-written response was outside their domain of expertise. A new internal guideline, reviewed by TechCrunch passed down from Google, however, has said responses can no longer be skipped and instead a note should be included that they don’t have knowledge on the domain. This new guideline has led to concerns that Gemini could be more likely to release inaccurate information on sensitive topics, such as healthcare.