Skip to main content

Trying the Google Bard AI chatbot : nice writing, but not entirely accurate (and a bit boring)


Today I got access to Google's chatbot named Bard, and I've been giving it a spin. 

According to the Bard FAQ, Bard is based on Google's LaMDA (Language Model for Dialogue Applications) conversational AI model. Google is clear that they did not use data from Gmail or other private apps for training.

Google also is very clear that it is not designed to return facts, help you code, or tell you about itself. Instead it is promoted as a creative tool that can help you brainstorm ideas for a party or write a draft email. As Google notes:

LLM [Large Language Model] experiences (Bard included) can hallucinate and present inaccurate information as factual.

When you sign in to Bard for the first time, there is a popup that makes it clear: 

Bard is an experiment

As you try Bard, please remember:

Bard will not always get it right
Bard may give inaccurate or inappropriate responses. When in doubt, use the "Google it" button to check Bard's responses.

Bard will get better with your feedback
Please rate responses and flag anything that may be offensive or unsafe.

All Large Language Models "hallucinate", including ChatGPT and Microsoft 365 CoPilot, which is touted as being "usefully wrong". 

My current impression is that Bard can be used for brainstorming, but it mostly comes up with nicely written, but pretty basic, suggestions. The information it includes is mostly accurate.

I found it most useful when I asked for cooking suggestions and recipes, as that information can be difficult to wade through in the search results. And that works for me because I know enough about cooking to start with the suggestions and do my own thing, rather than expecting to follow a precise recipe (which may not work). 

But in most cases the results to my prompts weren't bad, but were pretty boring. Boring isn't necessarily bad, of course. Boring is probably good for Gmail and Google Docs drafts, which is one of the first places Google is integrating generative AI.

I've also used Google's AI Test Kitchen, which is also based on LaMDA, and found the "Imagine It" app to be much more fun and interesting to play with as a creative tool.

And that is partially because "Imagine It" isn't a one-and-done response to a prompt. It asks questions and lets you refine your story. 

 This is presumably early days, so I'm interested in seeing where this goes. I'm hoping that future versions incorporate a back-and-forth discussion options, or at least the ability to refine prompts based on existing responses. 

Read on to learn how to sign up for Bard, manage what is saved in your Google account, and the result of the prompts I tried.

Sign up to try Bard

Currently Google Bard is still an "Experiment", and only available for testing to users with personal accounts that meet the following criteria: 
  • In the United States and UK. 
  • At least 18 years old.
  • Not using an account managed by a parent or guardian.
  • Not using an account managed by a Google Workspace administrator. 
 You can sign up to try Bard at bard.google.com.

Using Bard



Bard response to prompt "What are the elements of a good blog post".

Once you have access to Bard, all you need to do is sign in and enter a prompt. 


That means you agree not to "input any personal or sensitive information, including names, phone numbers, addresses, emails, or birth dates." 

And not surprisingly, it is not allowed to use Bard (or any other generative AI services from Google) for anything illegal, spamming, sexually explicit content, spams and scams, harassment, misleading claims of expertise, or even claiming the content was written by a human. I've written up an overview of the prohibited use policy here

Bard will generate 3 "draft" responses. You can give feedback on each response, and there is a handy "Google This" button to do an actual Google search for more information and get linked references.

Responses are customized by your past interactions with Bard and your general location (based on IP address). 

My first attempt was a question about myself, which is probably not a fair test. It's also, I realize now, a violation of the terms of service, as names are considered personal information along with phone numbers, addresses and other personal details. 

My question about Peggy K (the name I mostly use online) and it got some details correct. I did not include any details about my location or who I actually am (as that is likely considered personal information), but the description of me as a Google Product Expert was correct, with some flattering descriptions thrown in. 

Some information was completely wrong. I am not a yoga instructor. And, sadly, my YouTube channel does not have more than 10,000 subscribers. 


If I enter a similar prompt with my full last name, it was encouraging that Bard stated it didn't have enough information for a response. I would much rather it do that, rather than hallucinating information or providing personal details gleaned from search results. 
I do not have enough information about that person to help with your request. I am a large language model, and I am able to communicate and generate human-like text in response to a wide range of prompts and questions, but my knowledge about this person is limited. Is there anything else I can do to help you with this request?
But I figured brainstorming and suggestion prompts are probably a better test of Bard's capabilities.


Bard responds to prompt: "What can I prepare with carrots, onions, beets,
 peas and potatoes?" with reasonable-looking recipes.


Bard responds to prompt: "Where should I go for a hike?" with localized
 suggestions in the San Francisco Bay Area, which is the general area I am in.


Prompts I tried:
  • When I asked "What should I cook for dinner?" it gave me a list of 5-ingredient dinner recipes.
  • When I asked "What can I prepare with carrots, onions, beets, peas and potatoes?" it gave me three actual recipes that I think would work. At least I know how to cook well enough that I could use these as a starting point.
  • When I asked "Where should I go for a hike?" it listed four places to hike in the San Francisco Bay Area (my general location) with descriptions that looked accurate.
  • Asking "Are there any current road closures?" listed supposed Bay Area highway information, but it was not accurate. And by not accurate I mean it included a location that doesn't exist. 
And for brainstorming purposes, I asked "What topics about YouTube, Blogger and AdSense should I make tutorials about?". The result was fine - "How to create a YouTube channel", "How to earn money from a Blogger blog" - but boring. 


Bard's response to prompt "Write a paragraph about the latest changes to YouTube policy".
Clicking the Google It button makes it easy to search for related topic "Latest youtube policy changes."

The best result I got was from the prompt: "Write a paragraph about the latest changes to YouTube policy".  All the drafts correctly described YouTube's updated profanity policy for monetization, which was announced two weeks ago. That is correct current information. 

But it is still clear that Bard does not "understand" what it is writing, as it states "the new policy goes into effect starting today." It's clearly just grabbing text from another site or hallucinating common related phrases.

What about creative prompts? 

I asked Bard to "Tell me a science fiction story set 1000 years in the future". The result is fine, and might be useful as a starting point for crafting your own story. But it isn't as interesting as AI Kitchen's "Imagine It" process that lets you be part of the creative process. 


The bottom line: Bard can be useful for generating suggestions and the text is fine English. But I don't feel like it's a great discovery or brainstorming tool using most of the prompts I tried.

One feature I think would make Bard more useful would be the ability to iterate on a prompt. That would feel more like actual brainstorming or creating. I'm hoping that is included in future versions.

How does Google use your interactions with Bard?

When you interact with Bard, Google collects the following types of information: 
  • Your conversations (don't submit prompts with personal details!)
  • Your feedback
  • Your usage information
  • Your general location based on your IP address
Google makes clear that your conversations are not being used for advertising purposes.

You can manage how your Bard interactions are stored in your Google Account's activity settings. 

Visit myactivity.google.com/product/bard to turn off storage of your Bard activity (noting that it is always retained for 48 hours), or to remove specific activities (like prompts or feedback) from your Bard history.  

More Information

Comments

Post a Comment

Spam and personal attacks are not allowed. Any comment may be removed at my own discretion ~ Peggy