Google has opened up access to its AI chatbot Bard to a limited audience. “It’s an early experiment that will allow users to collaborate with generative AI,” Google executives Sissie Hsiao (product manager) and Eli Collins (research manager) wrote in a blog post Tuesday.
A factual error in the first demo of the Google text robot Bard has already caused ridicule and anger on the internet and caused Google stock to be lubricated.
“We’ve already learned a lot from testing Bard, and the next important step for improvement is getting feedback from more people.”
Who is Bard available to?
Google’s AI chatbot is initially only available to users in the US and UK and only in US English as part of a beta test.
The minimum age is 18 years. And you have to be on a waiting list at bard.google.com. However, this is not yet possible with a local Google account.
What is the first impression?
In a demo for The Verge, Bard is said to be able to quickly and fluidly answer a series of frequently asked questions, offer helpful advice such as how to get a kid to bowl (“Take them to a bowling alley”), and provide recommendations on movies and software.
In fact, there are only a few meaningful videos on the performance of Google’s AI chatbot available.
First impressions of the tester comparing Bard to Bing (Microsoft’s search engine has integrated the latest AI version of OpenAI, GPT-4):
- Bard responds faster than Bing to user input. This may be because the Google AI has far fewer users at the start of the public testing phase. In addition, the language model must be less than GPT-4.
- The normal Google search is more reliable for geographic queries (where are flower shops near a given location). In fact, a link to Google Search is also displayed below the AI text output, which struck the tester as rather odd.
- Google Bard failed on specific language questions around grammar, style, and spelling
- With relatively simple math (percentage calculation), Bing was significantly “smarter” than the Google challenger.
- When writing poetry, Bard seemed relatively dull. Bing is clearly more “original”.
- Google Bard and Bing (GPT-4) failed simple questions about historical data, such as how many days passed between the opening of the Eiffel Tower (March 31, 1889) and the Statue of Liberty (October 28, 1886).
- Google Bard tells much nicer jokes than Bing, but has trouble recognizing (user) humor.
- On the other hand, Google Bard was better at formulating so-called “prompts”, which are text inputs for AI software, in this case for artificial image generation. According to testers, Midjourney produced the more compelling graphics with the Bard prompts than with the Bing prompts.
As with ChatGPT and Bing, a note appears below the main text entry field: it warns that “Bard may display inaccurate or offensive information that does not reflect Google’s views.”
Is the AI hallucinating?
Yes. The same problem occurs with ChatGPT and other AI chatbots based on a similar language model. The software fakes answers. And so smart that it can be dangerous.
According to The Verge, Google Bard couldn’t correctly answer “a tricky question about the maximum load capacity of a specific washing machine,” but instead came up with three different but wrong answers. Repeating the question eventually brought up the correct information, but as a user you should have consulted a reliable source such as the manufacturer’s manual to see whether the information provided by the AI was correct.
The Verge further reports:
Why is this taking so long?
Google has had a leading position in AI research for years, but was afraid to open its systems to external access before.
With Bard, Google competes against ChatGPT from the California company OpenAI, which in turn is closely linked to the software company Microsoft through billions of investments. ChatGPT has caused quite a stir since its release (in November 2022) and has attracted a lot of public attention to the topic of artificial intelligence.
How similar are Bard and ChatGPT?
Like ChatGPT, Bard is based on a Large Research Language Model (LLM). OpenAI works with GPT, currently in version 4 (called GPT-4). This is also integrated into the new version of the Microsoft search engine Bing
At Google, Bard runs on a stripped down and optimized version of the language model LaMDA and is intended to be updated over time with newer, more capable models. Interested parties can register directly at the web address bard.google.com.
Hsiao and Collins continue to write that Bard is a direct interface to a large language model. “We see it as an addition to Google Search.” Bard is designed so that you can simply use the search to check the answers or explore resources on the web.
If you click on “Google it”, you will see search suggestions. Search opens in a new tab so you can find relevant results and dig deeper.
Google’s two executives promised to keep improving Bard and adding new features, including coding, more languages, and multimodal experiences. And one thing is certain: «We will learn with you as we work. With your feedback, Bard will continue to improve.”
The Verge’s preliminary verdict: Google’s AI language model LaMDA is much more powerful than the previously available, limited interface would suggest.
Sources
With material from the SDA news agency
- bard.google.com: Frequently Asked Questions about Bart
- blog.google.com: Try Bard and share your feedback
- theverge.com: Google is opening early access to its ChatGPT rival Bard – here are our first impressions
- support.google.com: Join the waiting list for Bard
(dsc)
Source: Watson
I’m Ella Sammie, author specializing in the Technology sector. I have been writing for 24 Instatnt News since 2020, and am passionate about staying up to date with the latest developments in this ever-changing industry.
