A First Glance at Google Bard
Unless you’ve lived under a rock, you’ve probably heard much about Chat-GPT. Well, now it’s time to look at one of its potential rivals: Google Bard or Bard AI. I started seeing advertisements that Google was doing an experimental preview of Bard AI. Naturally, I was curious, so I put myself on the waitlist. If you’re curious, I’ll leave a link to Bard AI at the bottom of the article so you can sign up as well.
What is Google Bard?
At the time of writing this, Bard is simply an experiment based off Google’s LaMDA and AI Test Kitchen projects. To get a better understanding for how we got to Bard let’s have a look at these two projects individually. I think it’s interesting to see where Bard came from, but if you’re not interested in the history feel free to skip to the next section.
Google LaMDA
LaMDA was released around 2 years ago and was the culmination of years of research in trying to get machines to understand search queries. LaMDA stands for “Language Model for Dialogue Applications” and, as the name suggests, is meant to carry on the flow of a conversation. The point of LaMDA is to be able to pick up on all the nuances of a conversation and be able to carry on that conversation with a person. We can already see how this would be helpful to something like a generalized conversational AI model. Rather than training LaMDA on lots of random text, Google focused its training data on dialog. I won’t get too far into LaMDA but if you’d like to read more you can do so here: https://blog.google/technology/ai/lamda/
Google AI Test Kitchen
Google’s AI test kitchen is an app that allows people to interact and give feedback on a variety of AI experiments, including LaMDA. It was a way to do a small and safe rollout of this technology and gain feedback on all the different experiments that Google wanted to run with it. This was released last year and Bard seems to be an extension of this test kitchen. You can read a bit more about Google’s AI Test Kitchen here: https://blog.google/technology/ai/join-us-in-the-ai-test-kitchen/
What to Expect with Bard
I always like to read up a little about what technology I’m using. Understanding where something came from gives me a better understanding and appreciation for what is in front of me. It seems like Bard is built on top of LaMDA as well as having some improvements based on the feedback from the AI Test Kitchen. Given that LaMDA was meant to be more conversational and pick up queues people give during the normal flow of a conversation, I am expecting it to be more like talking to a person than searching information.
Here’s my first interaction with Bard. You can see it’s trying to pick up on what I’m saying in the conversation. It gives typical responses to greetings. What I really wanted to know was how it would respond to me when I told it I had a long day. Most of the time, when someone says that, they typically mean they were stressed out, so Bard took that a queue to give me some de-stressing tips. I was also curious how it would respond if I told it that it had not been a stressful day as it had assumed. Overall, with this short interaction, I can see how they trained it to carry the flow of a conversation. One thing I assumed it would do is try to keep me talking and interacting with it. However, after my rebuttal of its assumption, it seemed to give what I would consider, something of a conversation-ender.
I tried to get it to continue the conversation by asking for a recommendation of the tips it had given earlier. This was a test of a couple of things.
Can it recall information from a previous message ?
Can it recall the context of a conversation?
Will it just parrot back the tips it gave me earlier?
I was pretty happy to see that it gave me a single recommendation as well as list of why it chose that one. Let’s move on and test its creative abilities.
How Creative is Bard?
Bard is advertised as a creative collaborator so I wanted to see how creative Bard can help me be. Since Bard recommended music, let’s put its musical chops to the test and ask it to write me a song.
To be honest, I don’t think this is a very good song, but it did come up with a song for me. Let’s see what else it can do. I love horror stories, so let’s see if Bard can help me create one.
I wanted to see if it will help me come up with a prompt for my story and to see if it would finish my story based on a limited amount of information.
I didn’t really get the feeling that Google understood what I was going for, but maybe I wasn’t explaining myself very well. I was trying to get it to come up with some subtlety about a pen that might make the pen in some way, shape, or form scary. Like if they were just stealing things and throwing stuff into a bad and took what seemed to be a standard pen, but later they find out it’s a human finger or something like that. I didn’t want to spoon-feed it the main point since I wanted to see if it’d come up with something like that on its own. Instead it gave me something more like this.
Either way, I think that Bard could be pretty useful as something to start bouncing ideas off of. It will continually invent new information that you may be able to make use of.
Strengths and Limitations
At the time of writing, Bard is still an experiment so it’s certainly not going to be perfect. One of the really cool things about Bard is that it has access to Google Search.
It’s certainly correct! You can verify that here: https://blog.angular.io/angular-v15-is-now-available-df7be7f2f4c8
In my mind, this gives it a slight edge over other language models. I’m not going to do a full comparison here but to give an example ChatGPT only has knowledge up until September of 2021.
This means that you can interact with Bard more like you’d interact with a search engine. Let’s see if it can summarize yesterday’s news (yesterday being Tuesday, March 21st, 2023).
Some of these stories were indeed reported on today. I decided to type one of the headlines into Google search to see what we get.
Sure enough, that’s today’s news. Aside from the fact that a lot of the news is quite depressing, the capability is actually really odd. It can ascertain today’s news more accurately than it can yesterday’s news.
While this article has been largely focused on what Bard can do well. I think we should talk about what it doesn’t do well. In this case, what it’s straight up missing. That is the ability to write code. This puts it far behind ChatGPT in terms of capabilities. To be honest, given Bard’s history I don’t find it surprising that it lacks that ability. It’s based on a model trained for dialog. I don’t know about you but my typical Friday night out doesn’t involve writing code back and forth with my friends in a group chat. Apparently, the Google team is working on integrating that feature, but we’ll see how well it can actually do it. Once Bard learns to code, I’ll certainly be back with another review of it.
I would encourage you to sign up and give Bard a test. It’s always interesting to see how different teams solve the same problem in different ways. I have enjoyed interacting with Bard and I think if Google can iron our some of the creases it’ll certainly be a strong challenger to ChatGPT!
Sign up for Google Bard: https://bard.google.com
