Google announces Bard: an AI-powered ChatGPT rival that will soon be featured in search

AI-powered chatbots may be the next big thing for Internet search engines. Microsoft certainly believes so and recently invested $10 billion in OpenAI, the makers of ChatGPT. Microsoft just announced that the AI model will be integrated into Bing and Edge, the company’s search engine and browser that are so far well behind Google’s alternatives in terms of market share.

Google has similar plans, though it will use its in-house research. CEO Sundar Pichai introduced Bard, an “experimental conversational AI service”. It will be integrated into a core Google product, search, and will first be available to “trusted testers” before becoming available to the public at large “in the coming weeks”.

Bard can answer questions directly instead of just pointing you to sites that might have the answer

What can you do with Bard? Instead of typing in keywords to find a site that might have an answer to your question, you can just ask the AI. It will pull fresh information from the web, using multiple sources and distill it into something easy to read.

For example, if you want to learn to play a musical instrument, you can use Bard to help you decide which one – would a guitar be easier or a piano? How much practice do you need? Multiple sites provide multiple answers, which Bard will summarize in a few easy to read paragraphs.

There is a question of how much you can trust its answers. For example, someone pointed out that the information from the video demo below is wrong – the JWST did not in fact take the very first picture of a planet outside of our own solar system. A quick Google search reveals this article from 2008, which says that the Hubble telescope took the first photo of an exoplanet (and explains why some even earlier claims of such photos are not widely accepted).

Bard is built on Google’s LaMDA, short for Language Mode for Dialog Applications, more specifically a lightweight version of LaMDA. This is smaller than the full model, which means that it requires less computational power to provide answers – something that will be important when Google tries to scale Bard to the point where it can answer questions from millions of users.

As for accuracy, Google’s internal team and soon the testers will be providing feedback to make sure the answers you get from Bard are safe and grounded. You can read this post from last year, which describes the key objectives – Quality, Safety and Groundedness – and how they were achieved through training and fine-tuning.

Google will use its AI tech in its own products, but starting in March it will invite individual developers, creators and enterprises to join in and built new things using the Generative Language API, which will (at least initially) be powered by LaMDA.

In other words, there will be a Cambrian explosion of chatty AIs in the coming months and they could change how we browse the Internet. And not just the Internet, Pichai promised that more AI-related announcements are coming soon.

PS. if you’re wondering whether ChatGPT answers the question about the James Webb Space Telescope correctly, the answer is no. ChatGPT was trained on old data (around 2020), before the JWST launched, so it only talks about what the telescope could discover in future tense. That is part of the reason why Google is highlighting that Bard is using fresh data, meaning that it can provide up to date answers.

Source