First impressions of Bard: Google's generative AI

The AI ​​race has begun. This has led to interesting alliances, such as OpenAI and Microsoft, but has also forced some tech giants to introduce their own AI projects.

That was the case for Google at the end of March 2023, with the launch of Bard. It was first made available to Pixel Superfans and is now open to anyone on the waitlist. Let's see some first impressions of this AI tool from Google!

What is Google Bard?

Bard is a Generative AI designed to generate text from naturally written queries. Bard uses Google's own language model, LaMDA (Language Model for Dialogue Applications). It is built on top of Google's Transformer - an artificial neural network architecture first announced in August 2017, which was very intriguing at the time.

Bard works like this: First, open a web browser and visit the Bard homepage (currently, there is no app for AI chatbots). Then, ask the Bard whatever you want, as naturally as you want. For example, the article asks him to write a short poem about the love between a proton and an electron who love each other, or to tell the size of the sun.

Whether you need the answer to a math question or want the Bard to help you think about the next chapter of your novel, Google aims to make finding a solution as natural as asking a question. human. Essentially, the Bard is a wise man in your pocket, ready to answer any question at a moment's notice.

What was your first impression of Bard?

If you're using Bing AI to help you plan a trip or find a new dinner recipe, you may have noticed that it can be creative and deliver accurate content but is slow. That's a problem with other Generative AI platforms like ChatGPT, because it seems like you have to watch line by line as the AI ​​answers your questions in real time.

If competitors like Bing AI and ChatGPT are marathon runners, then Bard is a sprinter. Ask Bard a question and you'll not only get an answer in seconds, but two more versions. One can imagine things like this.

For immediate answers - such as settling a nightly argument at the pub - the Bard seems pretty ideal. But note that, despite the slow speed of giving answers, Bard is still not 100% accurate.

What's missing from Google's Bard?

Even in beta, Bard remains easy to use and flexible. But until now, it has not been able to produce 100% accurate results and prove those results.

When you ask Bard a fact-based question, you'll sometimes see the source of the answer or a button inviting you to Google your question. This is one area where Bing AI dominates its competitors.

When asking Bing AI the same question it asked Bard before, the answer received was not only on par with Bard's best answer, but also came with four sources it could use to confirm its results.

When you ask Bard something a little too technical (such as medical questions), it usually responds with something cute like "I'm just a language model". language model). But there's an interesting discovery that if you talk to it like a normal, non-technical person, you can sometimes get Bard to reveal information it wouldn't otherwise share.

For example, the author of the article asked Bard about symptoms of arthritis in greyhounds. It does not list the full range of arthritis symptoms in greyhounds, but does provide a list of reasons why a greyhound may be tired and resistant to exercise. If Bard is to be a Generative AI assistant for people, it needs to be ready for any guidance, especially technical guidance.

Google's Bard may be easy to use, but Generative AI chatbots need more to compete in this emerging market.

Isabella Humphrey

Update 30 March 2024