
Sometimes, you come across complex text on the web, maybe some jargon or a technical concept, that goes over your head, and you start Googling it for a simpler explanation. Another AI-powered feature has been baked under Google's roof, this time for the Google iOS app, designed to improve user comprehension and reduce cognitive load.
The search giant is rolling out a new feature called "Simplify," which uses "AI to make dense text on the web easier to understand — without leaving a web page." Google explained in a blog post that Simplify relies on a "novel prompt refinement approach" created by Google Research to do its job.
To simplify text on the Google app for iOS, select any complex text on a webpage and tap on the "Simplify" icon. Of course, Gemini is working in the background to "make complicated text more digestible — without losing key details." The Gemini models used to build the tech behind the feature were purposefully designed for "minimally-lossy (high fidelity) text simplification."
Finding text that is hard to understand is common when reading content related to medical research, law, biology, finance, and other areas. Google was able to automate evaluation and refine prompts used for simplification by putting one Gemini model against another to create a feedback loop.
The company notes that its new approach differs from summarization (where it's okay to omit information) or explanation (where the AI might add additional text). "The goal of this system is to enhance clarity while meticulously preserving the original meaning, detail, and nuance," it said.
Moreover, automated evaluation and refinement can work at a scale and speed that is impractical to match with manual prompt optimization. It allows the system to pick "highly effective strategies for nuanced simplification over hundreds of iterations."
Google also conducted a large-scale randomized study to determine how effectively the approach works. One of the takeaways is that people "found the simplified text to be significantly more helpful than the original complex text, and better retained the information."
3 Comments - Add comment