Short answer
Bard is an experimental conversational AI service that Google announced on February 6, 2023, built on its LaMDA language model and able to draw on information from the web for up-to-date answers. For now only a group of trusted testers can use it; Google says it will open Bard more widely in the coming weeks.
Follow-up question When will you be able to use Bard, and how should you prepare?
Yesterday, Monday, February 6, 2023, Google's chief executive Sundar Pichai announced Bard, the company's answer to the wave of interest in chatbots set off by ChatGPT. In a post on Google's blog, he describes Bard as an "experimental conversational AI service" powered by LaMDA, a language model Google has been developing for several years.
Bard is not available to the public yet. Google is opening it first to "trusted testers," a group of outside users chosen to try it and report problems, and says wider access will follow in the coming weeks.
What Bard is
Bard is a chatbot: you type a question or request in plain language, and it answers in conversation, remembering what you said earlier in the exchange. Google says its goal is to combine the breadth of knowledge on the web with the creativity of its large language models.
A large language model is a program trained on enormous amounts of text to predict the next word in a sentence. Doing that well across billions of examples teaches it grammar, facts and styles of writing. Google's model here is LaMDA, short for Language Model for Dialogue Applications, which Google first showed off two years ago as a system built specifically for conversation.
Google gives two examples of what Bard can do. It can take the latest findings of NASA's James Webb Space Telescope and put them in words a 9-year-old would understand. And it can tell you about the best football strikers playing today and then suggest drills to improve your own game.
How it differs from ChatGPT
The biggest difference Google highlights is fresh information. Bard draws on the web to give current answers, while ChatGPT answers from what it learned during training, and OpenAI says its base model finished training in early 2022.
| Feature | Bard (as announced) | ChatGPT (today) |
|---|---|---|
| Maker | OpenAI | |
| Underlying model | A lightweight version of LaMDA | A model from OpenAI's GPT-3.5 series |
| Information | Draws on the web for fresh answers | Based on its training data |
| Who can use it | Trusted testers only | Anyone with a free account; $20 Plus plan in the US |
Google is starting with a smaller, lighter version of LaMDA. Pichai explains that it needs much less computing power, so Google can serve more people and collect more feedback. The company will combine that outside feedback with its own testing to check that Bard's responses meet its standards for quality, safety and "groundedness," meaning answers that stay tied to real-world facts.
The rest of the announcement
Bard was only one part of the post. Google also previewed:
- AI features in Search. For questions with no single right answer, such as whether piano or guitar is easier to learn, Search will soon show AI-written summaries that pull together several perspectives, with links to explore further.
- A developer API. Starting next month, Google plans to let individual developers, creators and businesses try a Generative Language API, first powered by LaMDA, with other models to follow.
- Partnerships. Google mentions cloud partnerships with AI companies Cohere, C3.ai and Anthropic, the last announced the previous week.
Why it matters
Web search is Google's core business, and the company has long presented itself as a pioneer in AI. Pichai notes that the Transformer, the network design Google described in a 2017 paper, now underlies many of the generative AI apps people are using. Yet until now, Google had not offered a public chatbot to rival ChatGPT.
CNBC, reporting through NBC News, says Google had been testing a chatbot called "Apprentice Bard" with employees as part of a "code red" response to ChatGPT. At an earlier all-hands meeting, Google's AI chief Jeff Dean told staff that Google faces more "reputational risk" than a small startup if it gives wrong information, and so it was moving more carefully. Bard's announcement shows Google is now ready to test in public, while still holding back a full launch.
Try it: Write down one question where up-to-date information matters, such as "What are the newest findings from the James Webb Space Telescope?", and one where it does not, such as "Explain how a telescope mirror works." Ask the second one to ChatGPT today, and save the first to compare when you get access to Bard.
When will you be able to use Bard, and how should you prepare?
Google has not given a date. It says only that Bard will become more widely available in the coming weeks, after the trusted-tester phase. Expect a gradual opening, possibly with a waitlist, rather than a single launch day.
In the meantime, the skills you are building with ChatGPT carry over directly. Clear context, a specific request and a requested format work with any chatbot. For example:
Explain what a black hole is to a curious 9-year-old. Use one everyday comparison, three short sentences, and end with a question they could ask a parent.One habit is worth strengthening now. A chatbot that draws on the web can be more current, but current is not the same as correct. When Bard arrives, check where its information comes from, compare important facts with a trusted source, and treat its answers as a starting point for your own reading.
Sources
- An important next step on our AI journey — Google (The Keyword, Sundar Pichai), February 6, 2023
- Google announces Bard AI in response to ChatGPT — NBC News / CNBC (Jennifer Elias), February 6, 2023
- ChatGPT: Optimizing Language Models for Dialogue — OpenAI, November 30, 2022