The Low-Down: Google's New AI Model 20 Percent Faster Than Chat GPT, Shows Sources

Google's latest Gen AI model is faster than ChatGPT's and can handle more content.

But perhaps most importantly - given the challenges AI faces due to questions about intellectual property violations and hallucinations - the Google model provides sources with a link to improve trustworthiness. JL

Laura Bratton reports in Quartz:

Google said in a press briefing that its latest AI model, Gemini 1.5 Flash, is 20% faster than OpenAI’s newest bot, ChatGPT-4o. customers are turning to Google in part due to one particular new update to its AI: accelerated “grounding” abilities, announced Thursday. Gemini models now show answers grounded (e.g. attached) to a source with a link. Those answers are also given a grounding (e.g. trustworthiness) score. Google said it will also launch a new industry-specific grounding tool in the third quarter, which will allow financial analysts, for example, to ground their AI queries against Moody’s data or legal experts to ground questions against Thomson Reuters sources.Goo
Google said in a press briefing that its latest AI model, Gemini 1.5 Flash, is 20% faster than OpenAI’s newest bot, ChatGPT-4o. Google announced Thursday that Gemini 1.5 Flash is now available to the general public.

The AI chatbot was first announced at Google I/O in May and has been available in public preview — meaning customers have been able to test the product and provide feedback — for the last month. Gemini 1.5 Flash can analyze one hour of video, 11 hours of audio, or more than 700,000 words in one query, rather than users having to break their questions up into chunks. In a presentation to journalists, Google showed how the bot could analyze a 14-minute video in one minute.

Google said the latest AI bot is also 40% faster than ChatGPT-3.5-turbo and has a higher capacity for information. In other words, users can include a lot more context — whether it’s audio, video, code, or text — in their questions for Gemini 1.5 Flash and, in return, apparently get more accurate answers.

“It’s the fastest model at the best price to performance option in the market,” said Google Cloud CEO Thomas Kurian. Gemini 1.5 Flash is available for developers in a limited capacity for free and, beyond that, at variable prices depending on how much data users want to input.

Gemini 1.5 Pro is about 10 times as expensive but described by Google as its “overall best model.” It can take in even more context: 22 hours of audio and 1.5 million words. “You can reason across an entire company’s history, which could be 10 years worth of financial statements” in one query, said Kurian.

Google stated that its recent updates to Gemini 1.5 Flash, its flagship model Gemini 1.5 Pro, and its image-generator Imagen 3 make its AI software bundle “the most enterprise-ready generative AI platform” on the market. It named several enterprise customers using its products, including UberEats, Moody’s, and Shutterstock.

Those customers are turning to Google in part due to one particular new update to its AI: accelerated “grounding” abilities, announced Thursday. Gemini models now show answers grounded (e.g. attached) to a source with a link. Those answers are also given a grounding (e.g. trustworthiness) score.

“...these elements are designed to improve factuality, substantially reduce hallucination.” — Google Cloud CEO Thomas Kurian on the company’s new fact-checking, or “grounding” features.

Moody’s Analytics Chief Product Officer Nick Reed told reporters, “Grounding is a pretty fundamental process in being able to use generative AI for decision making processes, particularly in financial services.” The financial services company is a customer of Google Gemini 1.5 Flash.

Google said it will also launch a new industry-specific grounding tool in the third quarter, which will allow financial analysts, for example, to ground their AI queries against Moody’s data or legal experts to ground questions against Thomson Reuters sources.