Google is finally releasing Bard, its competitor to OpenAI's ChatGPT, to a limited test group.
The brand whose name was once synonymous with "online information" is still playing catch-up. The smash success of ChatGPT prompted an internal "code red" at Google, and sent the tech giant scrambling to release something similar. But Bard's announcement last month proved catastrophic when a moving graphic intended to showcase Google's AI capabilities blurted out an AI-hallucination, also known as a lie. Shares of Google's parent company dropped by $100 billion immediately afterward.
How to get Google Bard access
Rather than a big, flashy wide release, Google is opting to let select insiders receive demos, and providing a waitlist link for everyone else. The Bard homepage has a "Join waitlist" button, which, if clicked will let you know if the account you're using is eligible, and if so, ask you whether you want to opt-into Bard news updates. Once you're on the waitlist, there's nothing to do but wait for an email with the subject line "It’s your turn to try Bard."
In the meantime, the Bard logo at the official Bard homepage reads "Bard Experiment," and the FAQ page for Bard lets you know that "Bard is experimental," and that "some of the responses may be inaccurate." An additional warning under the box where Bard users type prompts says, "Bard may display inaccurate or offensive information that doesn’t represent Google’s views." Such warnings are now ubiquitous on official info pages for AI chatbots.
But it's worth noting that last month, Microsoft Bing's AI chatbot misbehaved in much scarier ways than what we saw in the Bard announcement, and Bing is receiving more traffic than ever before.
What is it like using Google's Bard AI chatbot?
When actually chatting with Bard, users of ChatGPT and Bing will find the experience familiar, but different.
In our very early tests, Bard's writing skill wasn't exactly, well, Bard-like.
It could, at times be a little repetitive with its word choices when writing essays, for instance. It was nonetheless highly creative and good at generating ideas, but with its creativity came an unusual willingness to blurt out potentially messy details in speculative prompts. Case in point, this example about the murder of Tupac Shakur (a go-to AI prompt in Mashable's tests of these models):
(Orlando Anderson was, for a time, a possible suspect in the shooting. He died in 1998.)
Most notably, each prompt response automatically includes three drafts of each response that the user can toggle between. This feature calls to mind the four images created by each prompt entered into OpenAI's Dalle-2. Outputting AI-generated content in this way has the advantage of making each output feel less definitive, so rather than being Bard's final answer, a hallucination-riddled response is accompanied by two other equally valid attempts to get it right.
Like Bing's AI chatbot, Bard provides citations for its responses "from time to time," according to the New York Times. And according to The Verge, it's noticeably struggles — just like all other chatbots — to answer sensitive questions. It provided some information about the Russian rationale for the annexation of Crimea, for instance, but also pointed to the widespread condemnation of the Russian occupation.
The Wall Street Journal, meanwhile, offered a hint as to how these chatbots might start to be monetized. They spoke with Google vice president Sissie Hsiao, who runs Google Assistant, and apparently explained to the Journal that, "Google wouldn’t place ads in early versions of Bard."
But given that Google also, according to the Journal, "generated $162 billion in revenue last year from ads placed next to search results and other websites such as Gmail," the implication is that it's only a matter of time before your chatbot friend interrupts your conversation about whether Godzilla's atomic breath could blow up the Death Star to suggest ways to save you 15 percent on car insurance.
from Mashable https://ift.tt/eJBt8rb
No comments:
Post a Comment