Google I/O happened last week, in case you didn’t know.
Admittedly, it’s less of a cultural event than Apple’s WWDC, but it contains just as many important announcements. And this year was no exception.
Amongst beating Apple to it with the Pixel Fold and the hype surrounding the Pixel 7a’s 64MP camera, Google’s focus was largely on one thing. Yes, the topic we’re all equally sick, scared and in awe of; AI.
Following the ChatGPT storm a few months ago, Google went into panic mode. And I mean that literally. They were taken by surprise and feared the threat a generative AI model could pose to their search engine dominance.
The company threw together a team to create its own in response and they managed it, to varying degrees of success, releasing Bard for free recently in 180 countries.
But now a flip seems to have been switched at Google. They were caught napping once and they’ll be damned if it happens again.
And so, like a greedy toddler presented with a box of chocolates, Google couldn’t resist gorging itself on a boatload of AI-related tech. Let’s have a look at what went down at I/O 2023.
The biggest AI announcement at Google I/O was Bard.
Their chatbot to rival ChatGPT existed before I/O but was only open to those on a waiting list. Now it’s available for free in the 180 countries Google has rolled it out in.
And? Well, to be honest, it’s not very good. I mean, it can do stuff like write an email but it is significantly less sophisticated than ChatGPT. Bard reads as less Shakespeare and more SparkNotes summary.
But, in fairness, Google has had much less time to refine its model than OpenAI, and I/O made it clear that this was just the beginning. As proof, Google introduced PaLM 2, their updated language model that will now power Bard.
PaLM 2 promises better logic and reasoning, as well as a fairly unique focus on mathematics. PaLM 1 was trained on 540 billion parameters and Google hasn’t said how many for PaLM 2, but I’m assuming it’s a lot more. GPT-4 is estimated at 1 trillion, just for reference.
As I say, it’s still not as good as ChatGPT. But AI will be crucial to how search engines and app usage evolve in the near future, so you can rest assured that Google will try as hard as it can to dominate.
As part of this effort, Google I/O saw the company partner with Adobe to add image generation to Bard’s capabilities. This will be a significant step once it happens because no AI company has combined two different functions in a single bot.
It is this final point, on multifunctional AI, that seems key to Google’s plan to differentiate its offering.
Codey and Sidekick: An AI Ecosystem
Yes, Google is gunning for it all. They not only want AI everywhere, but they want their AI to be everywhere.
Codey is Google’s answer to GitHub’s Copilot. Copilot is a chat tool designed to help coders with issues they may face when working or to give coding prompts. It doesn’t fully code for you yet, but it undoubtedly will soon.
Codey is basically exactly the same thing. It too will use PaLM 2 to interact in a more natural and, dare I say it, human way with users. Ultimately, this illustrates how Google is seeking to maximise its AI output from a single model.
Same goes with Sidekick. Again, running on PaLM 2, Sidekick will work alongside word processing apps to suggest better phrasing when users are writing. In something similar to Grammarly, Sidekick will eliminate errors and create more fluent work.
It’s at this point you’ve got to ask, if AI can write original stuff in the first place, and now correct errors and perform rewrites…what’s the point in humans? But that’s for another, more depressing article.
The eagle-eyed amongst you will have noticed a theme here. Not a lot of what Google announced is original.
Sure, Codey (it’s a good name, you’ve got to admit) is designed with future integration across all of Google’s workspace platforms in mind, but at its heart, it’s a copied idea. Sidekick too and foldable phones. The Pixel Fold wasn’t even second or third onto the market! And Bard? Currently a pale imitation of GPT-4.
So is there actually anything exciting here?
The answer is yes!
This one is actually really cool. MusicLM is a text-to-music generator, much like NightCafe’s text-to-image.
This means users can input text and the AI model will produce a short clip of music based on the prompt. And it’s not just simple stuff like ‘play the keyboard’. It can get quite complex. For example, you can ask it to make music that is based purely on descriptive words.
I had a play around and asked it for a load of different things like a ‘spooky melody for Halloween’ and it worked. Sort of. It’s not great. Brian Eno is not trapped in a room on the other end composing it all. Often you just get a jumbled mix of sounds. But most of the time you can tell what MusicLM was going for.
And that’s pretty incredible when you think about what is going on behind the scenes. An AI model is processing the words, understanding how they correspond to a database of sounds, stitching them together according to genre tropes and considering tempo and volume, all in a matter of seconds.
It’s also an interesting way for Google to go at the moment. There is a lot of controversy over copyright infringement in song imitations and AI-created remixes. Even entirely original AI songs uploaded to Spotify are blacklisted.
So throwing their weight behind this feels significant.
Finally, we come to Google’s home ground. Their bread and butter. Search engines.
It’s certainly the least flashy of Google’s AI announcements, but it will also be the most prevalent for the everyday user. Search Labs is an entirely new app for Chrome users that uses AI to make Google searches more efficient and informative.
For example, AI will pull out relevant information snapshots to quickly show search results. Links to further reading will be automatically shown. Opinions from experts and forums will be displayed when relevant to add more depth to searches without taking up users’ time.
Admittedly, this doesn’t sound the most exciting, but I am actually the most interested in Search Labs out of all the glamour at I/O. This seems like a genuinely useful innovation that will make using Google more streamlined.
Rather than replacing someone’s job, Search Labs will be a simple but effective tool to gather more rounded information as quickly as possible.
So there you go. At Google I/O they announced foldable phones, mind-blowing 64MP cameras and music that is created at the drop of a hat. And by far the most interesting thing was a glorified Ask Jeeves.
I honestly believe this by the way, I’m not just saying this to be ‘cool’. I believe AI is most useful when it is not intrusive. Chatbots are great and all but they require a lot of collaboration. They are essentially another coworker.
But when AI exists in the background, automatically helping out, it hits the utility sweet spot. Like AI in smartphone cameras; its there all the time but we never see it working.
I’m excited to see how this strand of AI integration develops. I think it’s where AI is heading once the novelty of projects like MusicLM wears off.
As such, I’m intrigued by how Apple will respond to this AI dump from Google. We’ll be watching intently as WWDC23 grows ever closer, so stay tuned for all of Mark’s analysis!