Google is well-known for its expertise when it comes to search engines, maps and other tools designed to help people access information online. But there’s much more going on at the company than meets the eye – particularly when it comes to its generative AI model Gemini. As reported by The Information, behind-the-scenes contractors are responsible for evaluating and improving this AI’s output; recently however a change in these workers’ guidelines has raised concern regarding accuracy in response for sensitive issues like health and finance.
Gemini stands out as an AI that goes beyond text analysis by being capable of understanding audio, video, images and text in various languages – more powerful than traditional AI models which merely analyze text alone! Not to mention Gemini being integrated with Google products such as Search, Gmail Calendar YouTube etc to answer users questions quickly and provide solutions.
Gemini achieves this through Google’s Imagen 3 neural network incarnation. According to Google, this new model offers significant improvement over its predecessors with “significantly more creative and detailed generation of images” while remaining more accurate; meaning Gemini will likely pick out appropriate clothing items for people, make sense of speech and create better captions for photos.
Gemini has seen advances, but much more work remains before reaching the same quality we expect from other Google products. One key element is consistency – if Gemini’s information varies widely in accuracy, users could quickly lose trust in using it for important tasks such as healthcare or finances.
TechCrunch reported on Google’s internal memo instructing contractors not to skip prompts that require expertise they don’t possess, instead instructing evaluators to rate which parts they understand while noting any gaps in knowledge they might encounter. Prompts may only be overlooked if key information is lacking or contains harmful content requiring special consent forms for evaluation.
These changes could be the result of Google’s push toward efficiency, in which prompts are evaluated within an acceptable timeframe. It remains unclear whether this efficiency outweighs the risk of inaccurate results for sensitive topics. Google did not respond to our inquiries about this report but in a recent blog post X’s Jeff Dean touted Gemini as their most advanced and capable version to date for producing more realistic images that boost user trust in service. They plan to release public preview of next-gen facial recognition software this month.