Google Announces Bard, a ChatGPT Competitor You are Not Allowed to Use...Yet
It is more scalable than ChatGPT, but not right now
Google has unveiled Bard, a conversational AI service powered by its Language Model for Dialogue Applications (LaMDA). Initially launched with a lightweight version of LaMDA, Bard seeks to provide fresh and high-quality answers to questions by drawing information from the web.
Google aims to bring the benefits of AI into its everyday products, starting with search, to improve the depth of understanding of information and turn it into useful knowledge more efficiently. Soon, AI-powered features in search will begin rolling out, which will distill complex information and multiple perspectives into easy-to-digest formats.
Is This News?
I debated whether to write something up on this. I could not use Bard because it is not generally available. This has been Google’s problem right along. It shows a little peek under the curtain or just tells you there is a curtain and describes its color and texture. Then, it leaves you with the knowledge it has something so spectacular they must be careful before releasing it to the world.
Do you remember when Gmail was in beta for five years? If you signed up before July 2009, you were using the beta version of the software. Google spent a great deal of time in 2022 assuring us that Blake Lemoine was mistaken and LaMDA was not sentient. Given that the singularity and robot overlords are not a near-term risk, can you just release LaMDA? Haven’t you had enough time to work on the safety filters? Isn’t this a company of engineers? To paraphrase Steve Jobs, “Real engineers ship.”
ChatGPT to the Rescue
Apparently, Bard (or is it THE Bard, you might recall THE Assistant of years gone by) is only available to Google’s trusted testers. I was one of those testers when I received early access to LaMDA through Google’s Test Kitchen app. Maybe I will be invited to try out Bard. But, since I could not use it today, I decided to have ChatGPT Plus write the story introduction based on Google CEO Sundar Pichai’s blog post. Everything above the line in the first two paragraphs is ChatGPT’s summary of the Google announcement. It did a pretty good job.
This is not about ChatGPT to my rescue. It is about the rescue of Google. LaMDA was previewed with three use cases at Google I/O in May 2022. In August, LaMDA was added to the Test Kitchen app. There were only three use cases allowed for testing, whereas ChatGPT is unbounded. However, you might notice that those LaMDA dates were six and three months before ChatGPT.
You might say that GPT-3 was in the market in 2020, and InstructGPT, the alignment capabilities that led directly to ChatGPT, launched in OpenAI’s Playground in February 2022. So, maybe Google was behind. I don’t think so. LaMDA is pretty good at what it does. However, every month Google delays and opines about the amazing capabilities of the product we can’t use, expectations rise. That makes it even harder to ship because it will never live up to the expectations and will be scrutinized compared to ChatGPT, which everyone has tried. Talk about an innovator’s dilemma.
Waiting for the Bard
If ChatGPT didn’t arrive in December, Pichai would not have issued his “Code Red,” and I don’t believe we’d be close to seeing LaMDA available for public use. The creator of BERT and Transformers and TPUs and so many AI innovations is looking like a laggard. It’s not a laggard, but it looks like one, and that is being felt inside Google.
However, even that would not have been enough to shake the beast. The recognition that users immediately saw ChatGPT as a search alternative (this was also true for a small band using InstructGPT earlier in the year) and Microsoft was about to introduce the technology into Bing was an earthquake. The realization that Google Cloud risked falling further behind Azure in hosting AI model training and runtime operation was an aftershock.
Note this statement from Pichai’s blog post:
Next month, we’ll start onboarding individual developers, creators and enterprises so they can try our Generative Language API, initially powered by LaMDA with a range of models to follow. Over time, we intend to create a suite of tools and APIs that will make it easy for others to build more innovative applications with AI.
Next month is March. That will be nearly three years after the introduction of GPT-3 offered APIs to developers. And there is no link for developers to sign up for access or a waitlist. It is going to take some time for third-party applications based on LaMDA to generate momentum.
Granted, I said last week that despite Google’s plan to introduce 20 new AI products this year, we are unlikely to see any of them prior to Google I/O 2023 in May. Google is on track in that regard. It just seems desperate to launch a blog post the day after an unpleasant earnings release to talk about how much AI innovation you have driven in the past and that new products will be available real soon.
Pichai did say that Bard will be available in the coming weeks (presumably to trusted testers) and that you are likely to see the fruits of LaMDA show up in search very soon. I look forward to it. Google has a lot of amazing AI products that have been unable to leave port. Engineers ship. I look forward to Google getting back to its roots.
Yes, I found it ridiculous that they would announce a tool but not allow us to test it. I mean, I am more a user than a technician, but after working with ChatGPT, You Chat and other tools like ChatSonic I don't necessarily feel any appetite for Bard. Since it comes from Google, there is a certain level of curiosity.