# What is Google Bard, and how can you try it now that it is opening up?

Date: 2023-03-22 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/how-to-try-google-bard/

**Answer:** Bard is Google's conversational AI, built on a lightweight version of its LaMDA language model, and on 21 March 2023 Google began letting people in the US and the UK in through a waitlist at bard.google.com. Google calls it an early experiment: it offers several drafts of each answer and a button to check them in Google Search, and it can still state wrong facts with confidence.

Yesterday, 21 March 2023, Google started letting the public try **Bard**, its answer to ChatGPT. Anyone in the United States or the United Kingdom can now join a waitlist at bard.google.com, and Google is admitting people gradually. The company first showed Bard in February, but until now only testers could use it.

Google is careful with its words. It does not call Bard a product or even a finished service. It calls it "an early experiment" for working with generative AI, and says the main reason for opening it up is to get feedback from many more people.

## What happened, and who is behind it

The announcement came in a post by Sissie Hsiao, Google's vice president of product, and Eli Collins, its vice president of research. Access is starting in two countries and in English, with more countries and languages planned over time. Google has not said when Bard will be available to everyone, and reporting suggests the rollout will be slow.

The timing matters. In the past few months OpenAI's ChatGPT became a household name, Microsoft added a chatbot to its Bing search engine, and just last week OpenAI released GPT-4 and Microsoft announced AI help across its Office apps. Google, whose researchers invented the Transformer design behind all of these systems, is now putting its own chatbot in front of the public.

## How Bard works

Bard runs on a **large language model (LLM)**, a program trained on huge amounts of text to predict which words should come next. Specifically, it uses a lighter, optimized version of **LaMDA** (Language Model for Dialogue Applications), a model Google first presented in 2021. Google chose the lighter version so it needs less computing power and can serve more people. It says Bard will move to newer, more capable models over time.

Google's own explanation is a good one for beginners. An LLM works like a prediction engine: given your prompt, it builds a response one word at a time, choosing among words likely to come next. If it always took the single most likely word, the writing would be dull, so a little randomness is built in. That is also why the same question can produce different answers.

Bard's screen reflects this. Here is what you will see when you ask a question:

| Feature | What it does | Why it is there |
|---|---|---|
| Drafts | Up to three alternative versions of the answer | Lets you pick the best starting point |
| Follow-ups | Keep the conversation going in the same thread | Refine or correct the answer |
| Regenerate | Ask Bard to try again | Get a fresh attempt |
| "Google it" | Suggests searches and opens Google Search in a new tab | Check Bard's claims against real sources |
| Exchange limit | Caps the number of turns in one conversation | A guardrail to keep chats on topic |

## Why it matters, and what Google admits

Google is unusually frank about the flaws. LLMs learn from text that contains real-world biases and stereotypes, and those can show up in answers. They can also be wrong while sounding certain. In its own announcement, Google shows Bard suggesting easy indoor plants and getting the scientific name of the ZZ plant wrong: it is *Zamioculcas zamiifolia*.

That candor tells you how to use Bard. Google describes it as a companion to Search, not a replacement. The Google it button exists precisely so you can verify what Bard says. Early reports also note that Bard answers quickly, perhaps because few people are using it so far.

For Google, the stakes are high. Search is the core of its business, and conversational AI could change how people look for information. Bard is the company's first big public test of that idea.

```prompt
I want to read more books this year. I have about 20 minutes a day and I tend to give up halfway through long novels. Suggest five practical habits, and explain why each one works.
```

> **Try it:** If you have access, ask Bard the prompt above, then open all three drafts. Compare them side by side: which one is most practical for you, and do any of them contain a claim you would want to check with the Google it button?

## How do you get the most out of Bard while it is still an experiment?

Start by joining the waitlist at bard.google.com with a Google account, if you live in the US or the UK. Access is being granted gradually, so it may take some time before you are let in.

Once you are in, treat Bard as a brainstorming partner. Google's own examples point to good first uses: tips for a personal goal, a simple explanation of a hard subject such as quantum physics, or an outline for a blog post. Open the other drafts before settling on an answer, and use follow-up questions to steer it, for example "make that shorter" or "give me examples for beginners."

For anything factual, such as names, dates, numbers or health and money questions, check before you rely on it. Press Google it and read the sources yourself. Remember that each conversation has a limited number of turns, so if a chat runs out, start a new one and restate the key details.

Finally, share feedback when Bard gets something right or wrong. Google has said plainly that feedback is the reason for this release, and that coding help, more languages and multimodal features (working with images as well as text) are on the way. What early users report will shape what Bard becomes.

## Sources

- Try Bard and share your feedback: https://blog.google/technology/ai/try-bard/
- Google starts opening access to its Bard chatbot in the US and UK: https://siliconangle.com/2023/03/21/google-starts-opening-access-bard-chatbot-us-uk/