Google is now offering Bard – its chat-driven rival to ChatGPT – to carefully selected netizens in the US and UK.
Bard, derived from the web advertising giant's large language model LaMDA, was built to compete with OpenAI's GPT series, the brains behind the chat-bot interface for Microsoft's Bing search engine, 365 suite, and other applications.
ChatGPT dominated headlines, and took the internet by storm shortly after it was made accessible to world-plus-dog for free last November. Reports that Microsoft would incorporate a ChatGPT-like conversational system into its Bing search engine set alarm bells off at Google, and CEO Sundar Pichai declared a "code red" emergency - ordering employees to quickly build its own rival AI web search chatbot. The idea being that rather than type in keywords to search the web for, you instead ask the bot questions in natural language, and it answers, drawing upon what it's learned from the 'net.
Now, Google is finally launching Bard weeks after Microsoft unleashed its AI-enabled Bing to millions of users around the world. Bard was previewed, with somewhat bittersweet results. Now Google thinks it's ready for the mainstream, ish.
"Today we're starting to open access to Bard, an early experiment that lets you collaborate with generative AI," Google's Sissie Hsiao, VP of Product and Eli Collins, VP of Research, announced in a blog post.
Google fans in the US and UK can now sign up to join a waitlist to use the system. Like the Bing bot, which was also made available via a waitlist, Bard is designed to be a conversational agent capable of responding to general questions with answers that may or may not be correct.
On Tuesday, Microsoft enabled image-generating capabilities in its new Bing bot using the multi-modal abilities of OpenAI's latest GPT-4 model.
Meanwhile, Adobe released Firefly, a suite of generative AI tools designed to help people create art. It claims this has been trained on its own images, not those of its users. We'll see.
Large language models are like prediction engines, the pair explained: "When given a prompt, it generates a response by selecting, one word at a time, from words that are likely to come next."
Since Bard is powered by LaMDA and responds to input queries by predicting what response is most appropriate.
As a non-intelligent, information regurgitation engine, it doesn't really know the answer to a question, nor understands the actual problem; it just draws from what it was trained on, which is mountains of data sourced by Google. And it can generate toxic text, make stuff up by getting its predictions horribly wrong, and spread inaccurate information, a property described as hallucination.
They can provide inaccurate, misleading or false information while presenting it confidently
"For instance," Hsiao and Collins said, "because [these kinds of bots] learn from a wide range of information that reflects real-world biases and stereotypes, those sometimes show up in their outputs. And they can provide inaccurate, misleading or false information while presenting it confidently.
"For example, when asked to share a couple suggestions for easy indoor plants, Bard convincingly presented ideas … but it got some things wrong, like the scientific name for the ZZ plant."
Since Bard isn't perfect, users will see a few different responses generated by the chatbot and can pick the best one to follow up with.
Google described Bard as a "direct interface" to its large language model and a "complementary experience" to Google Search. People should use Bard as a starting point when searching for information, and are encouraged to find more relevant sources on specific webpages, the biz said.
In the future, Google plans to make Bard run on more powerful and larger versions of LaMDA, as well as adding capabilities to generate code, images, and support for more languages other than English. ®
Source: The register