20 Google Bard Announcements. How to Access it Today in 180 [non-EU] Countries.
Extensions, images, new languages, and even some live demos.
Bard was a featured product today at the Google I/O annual developer conference. The ChatGPT competitor has been in a closed preview since February. As of today, it is available in three languages and 180 countries. Notably, EU member nations are excluded from access. No explanation was given, but the reasoning seems obvious if you have been following generative AI news.
Here are 20 announcements, clarifications, and new information about Bard revealed at today’s event. Also, number 20 on the list may be of particular interest, given recent ChatGPT feature demonstrations.
No more waitlist - The big news is that Bard is available to anyone with a personal Gmail address. You can access it here.
Bard is now available in 180 countries - You should be able to access Bard almost anywhere in the world today except European Union nations. OpenAI may have done Google a favor by going first. Its ChatGPT experience showed what will happen when you introduce an experimental app and make it available to EU citizens. EU countries will move to regulate the service as a production-grade consumer product. That may be the right approach for the EU. Google is showing that the risk of regulatory action means the right approach for some companies will be to limit access. EU citizens can use a VPN to try Bard or fire up ChatGPT or Bing Chat and stay in the OpenAI ecosystem.
Bard is available in three languages - English, Japanese, and Korean are supported today.
More languages are coming - Google says that Bard will expand into 40 more languages, though these were not listed, nor was a time frame announced other than “soon.” A graphic showed several languages, including Danish, Dutch, French, German, Hebrew, Hungarian, Italian, Norwegian, Polish, Portuguese, Romanian, Spanish, Turkish, Vietnamese, and several others.
Bard is not a search engine. It’s a chatbot. - This was not explicitly stated, but when you use Bard, you will notice the user experience is more like ChatGPT than Bing Chat. This became more clear that Google was explicitly separating the conversational AI chat app and search during the I/O segment on search product updates. There will be a new “Converse” button in Google search that will enable conversational search. That experience lets you switch between traditional and conversational searches by scrolling up and down, similar to Bing Chat.
Bard is an experiment with conversational AI - Google CEO Sundar Pichai introduced Bard as a “conversational AI experiment.” The Bard app says “experiment” beside the menu bar and logo. The message to users: “Don’t blame me if it doesn’t work very well.” Also, a disclaimer below the text entry dialog says, “Bard may display inaccurate or offensive information that doesn’t represent Google’s views.” Pretty standard for generative AI apps today and not new.
Improvements in math, logic, and reasoning - No evidence was provided to demonstrate these improvements, but this claim was made twice. Results may vary. 😮
Enhanced coding support - Some coding features were discussed last month, but new examples and information were shared at I/O. The announcements included support for 20 coding languages, the ability to explain code functions, suggest code improvements, and export to an IDE. This is Bard-specific. Google AI Duet was also announced as a coding assistant pair programmer, more like GitHub Copilot or Amazon’s CodeWhisperer. Bard can perform some of these features as well in the chat interface.
Bard is (supposedly) available in Google Workspace - If your company uses Google Workspace with Docs, Sheets, and Slides, your administrator should be able to activate Bard features in these apps. There is a convoluted support document about activating “experimental apps,” but nowhere to turn on Bard as a service. This supposedly became available as of May 5th, and maybe it will be more easily accessible soon. For now, start with a personal email address.
Export Bard results to Gmail and Docs and sheets - There is now two click export of Bard responses to Gmail and Docs.
Bard does tables - You can ask Bard to create a list with additional information about each item and present it as a table. You can also ask it to add more information and expand the table. Note how it didn’t get tripped up by my typo at the end of the request.
Bard table data copies nicely into Sheets - Bard offers the option to copy any response, and if you do this with a table output, it will paste nicely into a Google Sheet. This also works with Excel, but Google didn’t mention that. 😅
Visual responses (maybe) - Visual responses to prompts were demonstrated with both Bard and conversational search, but Bard cannot do this yet. It does provide brackets with an image description where a photo would be returned but no images. Google's Sissie Hsiao said, “Bard will become more visual,” using the future tense. Let’s assume this is coming soon. 🧐
Prompts with images - In the future, you will be able to upload a photo and ask questions about it. You will also be able to use an image as part of your prompt request. There was a reference to Google Lens, but suffice it to say this is not available today.
Generate new images with Adobe Firefly - This is not available today. Bard will respond that “I can’t create an image…” and it is not aware of the Firefly integration. But Google says it will be available in the future. More notable is that Google’s text-to-image AI model Imagen is now available for access via Google Cloud, yet for Bard, Google is choosing to use Adobe Firefly. The rationale here is unclear.
Integration with Maps - Google announced Bard integration with Google Maps, but I can only get it to generate “[image of Google Maps directions]” in the response. Let’s assume this is coming soon. 😎📍🗺
Dark theme - Bard offers both a dark and light theme. Let’s call this newish.
No more LaMDA. Bard now runs on PaLM 2. - The Google Bard you saw demonstrated in February and that I have been using for three months was originally based on the LaMDA large language model (LLM). Pichai announced that the current version of Bard runs on PaLM 2. This is an LLM with three times more parameters than GPT-3, though it is unclear how it compares to GPT-4 since that information is not public. What is clear is that LaMDA was determined to be inferior to PaLM during the recent Bard test period and was unceremoniously replaced.
Bard will move to the Gemini LLM - Google made a big deal about PaLM 2 powering Bard without once mentioning the LaMDA model. However, Google also mentioned multiple times that Bard would migrate to a new LLM called Gemini at some point in the future. Gemini is currently in training, and Pichai lauded its multimodal capabilities. I presume this was mentioned as a way to foreshadow a movement to a more powerful or better-aligned model and preempt criticisms about the output of the current model and shifting models again. Let’s see how that works out.
Bard Extensions - Google also announced Bard Extensions, which is how third-party software and websites will become accessible within the Bard UI. You would be correct to assume this is Bard’s corollary to Chrome Extensions and ChatGPT Plugins. You may also notice some overlap with current ChatGPT Plugins from Instacart, Kayak, OpenTable, Redfin, Wolfram, and Zillow.
Subscribe to Synthedia. It’s free!
Google I/O represented a much better showing for Bard than the February debut. Most of the highlighted features were shown as images or videos, but there were a few live demonstrations. Most of Bard’s features are coming soon.
Google clearly recognizes it is not yet on par with Bing Chat or ChatGPT. It is in catch-up mode. That is why there were so many disclaimers, endless comments about the company’s work on making AI safe, and references to the new, better, and not yet available Gemini LLM. However, at least you can now see a path to parity with today’s announcements.
I jumped into Bard again today to ask about LLMs. The answers it produced were adequate, except when it lost context and started telling me about Camelids instead of the Vicuna LLM. Bard also rarely provides standard source links in Bing Chat, though I saw my first one today with a link to Huggingface.com. Progress is evident.
However, the messaging around Gemini sums up where Google is with generative AI today. Google wants us all to look to the future to envision its contribution to generative AI. That future may be near, but I/O was more about what will be than what is.
That said, Google Bard is available today for all of you non-Europeans. Let me know what you think in the comments.