Have you heard of Google bard yet? Google bard is Google's answer to open AI's  chatgpt which has been one of the most trending topics on the internet. 

We at Blueguard were fascinated by this new announcement and looked deeper into the potential of Google bard and the technology behind it. 

But before I tell you what we found, make sure to share this post and be coming back to this blog regularly to be updated about advancement in the world of AI. 

Google Bard vs ChatGPT 4

Google bard didn't really start out as Google bard but as Google lambda 2. And lambda 2 was interesting but limited in capabilities. People were still excited about the new idea and were looking forward to the future of lambda 2. 

A few months later, OpenAI released a full suite of different products that blew people's minds. ChatGPT from OpenAI in particular was extremely good among this and served as a direct competitor to Google's lambda 2. 

The platform gained millions of users in just a few hours after its release and big companies like Microsoft integrated their apps with this technology. For example, Microsoft's new Edge browser along with Bing can support chatgpt-based searches. This feature by itself can rival Google AI and is definitely something to be considered in the AI arms race. 

They recently launched their beta test for trusted users. This essentially means that the software is going through private testing and meanwhile for the competition they have released a solution called Nora. Nora stands for no right answer. This is for the problems where you don't really have a single plausible answer and the answer can have multiple choices of right options. This focuses on improving user experiences while empowering creativity, contribution and the challenge of developing an open web. 

This software is a game changer for search experiences and as an example let me tell you that it provides a feature wherein you can combine multiple images to form a single 3D image and that is pretty done cool. They have also introduced features that can give you product recommendations, pairing suggestions and much more. This really is a full utilization of what generative AI is capable of. To be honest, Google does have an excellent product and the quality of the outcome would just depend on how well they develop it. 

Google still has one of the largest user bases and I think this will definitely impact how the product will perform. Only time will tell what Google has in hand for us but as of right now the most concerning factor is their lack of innovation for user experience. 

We're living through an AI revolution. In just the last few days, Stanford released Alpaca AI, an LLM that can be trained for $600, and then subsequently took it down for hallucinating too much. Bing and Adobe both announced their own AI image generators.

Microsoft Office introduced Copilot, but the biggest news is Bard is now available as a public beta. Another good news for Google, though, is that they have more data about search than anybody else, they have virtually unlimited computing power on their own custom hardware called TPUs, and a team of geniuses working on top secret projects like quantum computers. They even declared quantum supremacy, so I have extremely high expectations for Bard.

I compared it to GPT-4 using four different experiments, and the results are absolutely shocking. 

But first, what's with the name Bard? Apparently it's supposed to mean Celtic storyteller. If I were in charge, I would name it something like Neurodominus 9000, or Encephalon Prime. It should strike fear in the hearts of the Microsofties.

The first thing I wanted to find out is if this thing could write code, so I prompted it to build me a basic to-do app with JavaScript. Amazingly, Bard gets the entire job done in just a few seconds. It takes GPT-4 at least five times longer to complete the same problem. That's pretty impressive, but going fast isn't very useful if you're dumb. Let's find out if this code actually works. I ran the code in the browser and opened the console and immediately got an error. Basically, it's trying to add an event listener to an element that doesn't exist. I know what the issue is, but I want to see if Bard can fix it. I tell Bard it's not working and paste the error message into the prompt, and it does come up with a solution using a conditional statement that would actually fix the error, but it doesn't make the application work, because that element still doesn't exist. I try to nudge it in the right direction, but then it hallucinates and tells me to add an event listener to an input, which wouldn't work. I keep trying to nudge it, but then it fails catastrophically.

Mission failed. We'll get up next time. Now, if we look at the GPT-4 response, you'll first notice how it's much more well-documented. In addition, when I try to run the actual code, I get a perfect working to-do app with no errors. I ran this experiment multiple times on different demos and got similar results, so round one goes to GPT-4. 

 In round two, we'll see how creative it is when it comes to idea generation. I've always wanted to write a romance novel, so I prompted it for some ideas. The Bard response was pretty simple and cliché, like the wrong man. A man and woman meet and fall in love. Blah, blah, blah.

It's written at about an eighth grade level. GPT-4 has a lot more creative depth, like Stardust melodies. When a small-town musician Ella lands a gig at a prestigious jazz club in New Orleans, she never expected to meet Alex. A world-famous saxophonist struggling with his own demons. They journey through the vibrant history of jazz to create a love song that will last forever. Now that's a fucking romance novel. 

So GPT-4 can come up with better ideas, but who can write better poetry? The next prompt is to write a poem about programming and the style of Dr. Seuss. In this example, Bard gets the job done, with lines like, I do not like this coding prob, I do not like it, Bob the job. That's pretty good, and one thing that stands out here is that it also provides a source for this work, which links back to some guy's blog post, which provides a little more transparency into how it generated this work. Bonus points to Bard for doing that. But now let's read GPT's four-palm.

He'd build them with care, with love most of all, from the Bina Roo bird to the Algum rhythm ring of Roo. As you can see, they both sound like Dr. Seuss, but there's just something special about GPT. Each stanza just drips off the tongue like honey, and that brings us to round four.

 Spitting Facts, where I prompted them to explain how brainwaves work. Bard is way faster at outputting information, which is very useful for problems like this. I'm in the process of turning myself into a cyborg for my next post, so make sure you're subscribed, and I can tell you that both of their responses are accurate. I would say GPT4 is slightly better, but I'm going to give this round to Bard, because I'm kind of starting to feel bad for it. Bard is also getting flamed on social media right now, and some people are doing the unthinkable, switching from Google to Bing. And Bard itself is already paranoid that Google is going to kill it off in the next one to two years. .

Print this post