Google I/O 2023 Impressions: A Whirlwind of Announcements
Folding phones, large language models, and corporate irony
Just a couple months ago, Microsoft unveiled Bing Chat, their artificial intelligence chatbot powered by OpenAI’s GPT-4 model, putting Google into overdrive mode to make Bard, their LLM powered by LaMDA a couple weeks later. The truth is, Bard… sucks — Google employees have admitted that Bard is a failure of a product, have said that it’s a ‘pathological liar’ and flamed Pichai for rushing the product’s launch. Google isn’t happy about the way the launch of Bard went — it’s painfully obvious. Bard didn’t even know how to code at launch, had no dark mode, and had no integration with other Google services.
All of that changed at Google I/O this year — in addition to genuinely exciting hardware announcements, Google ‘upped the ante’ on AI — so much so that they said the word AI literally hundreds of times during the keynote. Google really wants to show that it’s serious about artificial intelligence — after all, Google is the company that has the potential to benefit the most from LLMs. Nobody uses Bing — they use Google. Google/Alphabet CEO Sundar Pichai took the stage at I/O to say that Google has four main objectives in regard to AI — 1. To improve knowledge and learning; 2. To boost creativity and productivity; 3. To enable others to innovate; and 4. To build and deploy AI technology “responsibly.” This is taken verbatim from Pichai’s presentation on the 10th, and it sends a strong message. Google is and has primarily been an AI company for many years — the Knowledge Graph is powered by AI, the YouTube “algorithm” is just… artificial intelligence — heck, Google’s main selling point for their Pixel lineup of phones is that they’re the “smartest” smartphones. Google is a master at using AI to do things; see: Google Duplex. But somehow, they’ve fallen flat in the race of generative AI powered by large language models. Google will be the first to tell you that they built the T in GPT — the transformer architecture. Why aren’t they doing anything with this power and technology? What’s holding them back?
Google upgraded Bard’s model to version 2 of their Pathways Language Model (PaLM) — a model more powerful than LaMDA, which was announced at Google I/O in 2021 following Meera in 2020. Google says that their main objective in doing so was to improve Bard’s reasoning and logic — Bard can already access the internet through Google and as a result, has a huge data set of (mostly) accurate information on hand when it needs it — where it falls apart is in understanding complex queries and breaking them down into smaller problems that it can solve. PaLM 2 is supposed to address this, similar to the way GPT-4 does. PaLM also fixes the issue of Bard not being able to code — while not well, Bard can do some coding work now, and it can even provide citations and more information if needed. These are important, well-deserved improvements, and Google seems so confident in Bard’s new abilities that they’re willing to let you try Bard’s tech out in search with a feature called SGE — which nearly replaces the ’10 blue links’ Google has had since its conception.
However, I’m not as confident as Google appears to be in Bard’s abilities — Bard still can’t answer questions with the accuracy, precision, and conciseness that ChatGPT (using the GPT-4 model) does. Take this very simple example — I asked both ChatGPT with browsing and Bard the same question: “When was the iPhone 14 Pro announced? What is the headlining new feature?” ChatGPT gave me a helpful response — it accurately told me when the phone was announced, and accurately stated that the headlining feature of the phone is the Dynamic Island. It provided sources, was straight to the point, and was correct — all in reasonable time. Bard, on the other hand, failed to give me a count of the phone’s headlining new features — it completely missed the Dynamic Island, and instead told me that features from years before were “new” with brazen confidence. This isn’t the most important test, nor is it particularly scientific — but it says something — that Google is falling behind in the LLM race. As much as these announcements from I/O look good on paper, they still don’t fix the big, underlying problem: Bard lies more than ChatGPT does, and that’s a bad look for Google. Especially if they plan on rolling out Bard’s features to the masses in search soon, which they’ll have to do if they’re keen on showing the market that they’re competitive.
Google knows they can’t fall behind — in addition to bringing improvements to Bard, they’re bringing generative AI to Google Workspace apps and Gmail similar to how Microsoft aims to do with Office 365 (or is it Microsoft 365, now?). I’m excited about this addition — it means less switching around between browser tabs, which is almost always a good thing. I’m interested to see how many people will actually use this feature versus just toying around with it — is Google’s AI trustworthy enough in the eyes of most people? The company says that the feature will first be rolled out to business and education customers, which is also a peculiar decision because it runs the risk of data privacy issues — do companies want their employees revealing confidential information in Google Docs (like company meeting transcripts, etc) and important business communication in Gmail to an AI chatbot that uses conversation history for training? I’m not so sure. And will management or school administrators be able to see which parts of a text were generated by AI? This raises many moral and ethical questions worthy of their own column — but the point is, how will AI transform work as we know it? Google doing stuff is massive — millions upon millions of people use Google Workspace apps to do their work. The ramifications of these new features are massive.
There’s also a small bit of comedic irony here in the way that these features were announced. Gmail has 2 new additions powered by LLMs: first, a feature that asks you to generally outline what you want to say in an email so that Google can expand it out into a lengthy email with a formal introduction and conclusion, etc., and a feature that allows you to summarize lengthy emails into a concise outline that’s easy to read. Both parties want brevity here — why are we making AI an inconvenient middleman? I just thought that this was remarkable — we have so much technology that we could use for so many amazing things, but we instead just make it play a cat-and-mouse game of summarizing things it wrote itself. It’s just an inconvenient feedback loop!
None of this is to say that AI isn’t going anywhere — it’s here to stay and it will get better. But Google’s LLM technology is falling behind, and that isn’t good. Google is the company with the money, resources, and tech stack to make generative AI succeed. They just aren’t doing it yet. You know something’s wrong when your own employees are mocking you for using the word AI too many times during your headlining software keynote. AI-AI-I/O, am I right?
Even though it may not seem like it, AI isn’t the only thing that Google discussed during the I/O opening keynote — Made By Google, Google’s hardware division led by Rick Osterloh, took center stage to unveil the Pixel Fold, a device leaked into oblivion for the past 2 years, along with the Pixel 7a, and Pixel Tablet — a device full-on unveiled one year ago at I/O 2022.
The Pixel Fold is my favorite of these announcements for a specific reason — I’m a massive believer in foldable devices. The Pixel 7 Pro is my favorite Android phone of 2022 for a reason — they’ve nailed the software down so well (coming back around to Google making the smartest smartphones). Add the amazing camera and robust build quality, and I think it’s a better all-around smartphone than Samsung’s S23 Ultra. Now, imagine all of that goodness… but folding! That’s even better — it’s 2x the fun, and 2x the utility… with a big if. If Google can get the software right, this phone is a winner. From the likes of MKBHD and MrMobile, we’ve seen that Google has done a lot to make the Android experience more optimized for foldables. There’s now a dock pinned to the bottom of the screen which appears to show recently used apps and shortcuts like a desktop computer would have. In addition, Google also spent a while during the keynote talking about how they’ve optimized their own apps, like Google Docs, Chrome, and Gmail, for the phone. Apps should now feel more desktop-like and behave better to fill up the larger screen and should have more options for split-screen multitasking. It’s tough to say if this software experience will be better than what Samsung has at this stage, but here’s hoping that Google has things thought out. I’m rooting for this phone.
Specs-wise, the phone is flagship in pretty much every way — it sports Google’s flagship Tensor G2 processor, which might not impress on paper, but handles well after Google’s software patches last year. Also, it supports wireless charging, which is nice, a 48 MP rear camera, 4K 30fps video, and the usual selection of biometrics — face unlock using the 10 MP front camera, or a fingerprint sensor located on the power button — all normal, and expected for Google’s flagship device. On paper, the Pixel Fold resembles the Pixel 7 Pro in many ways, which is mostly a good thing — aside from battery capacity. Google says the Pixel Fold has a 4821 mAh battery, which, granted, is higher than the Z Fold’s 4400 mAh battery, but knowing how the Pixel 7 Pro fairs in battery life with its 5000 mAh battery (which is to say, not very good — average, but not very good), I’m not optimistic about how good battery life is going to be on this phone. I would’ve excused its smaller battery if the phone had fast charging that was actually… fast, but it doesn’t — the Fold only charges at 30W wired through the USB-C port. We won’t know how good the battery is on this phone until reviews come out later this year, so it’s unfair to deploy concrete judgment at this stage, but I feel like that’ll be the only gotcha of this phone.
But once again, I’m rooting for the Pixel Fold — I’d go as far as to say that it’s my favorite looking of the foldables out on the market. I’m a massive fan of the aspect ratio of the Pixel Fold, specifically — most folding phones are tall to preserve a squared aspect ratio on the main interior display, but like the Oppo Find N2, the Pixel Fold has a smaller, more compact-looking front cover screen. It’s only 5.8 inches and is kind of… how do I put it… stubby? The phone is short and small, is the point, and I love it; short phones are simply just nicer to hold in the hand — they have that iPhone mini feeling that I love. This approach works perfectly in a folding device — the main screen is small, but when you want a bigger screen for watching content or playing something, the phone just expands into a tablet-like aspect ratio. Additionally, the Fold is relatively thin — about 1.5 Pixel 7 Pros, which makes it feel even nicer. Google did some clever engineering work here to achieve this — most foldables have all of the hinge flex connectors and such embedded within the hinge, creating a gap between the two halves of the phone when folded. Google subverted this by putting all of the electronics at the top and bottom of the hinge and covering them using the display bezel. The result is a remarkably flat and thin phone that’s also water resistant by IPX8, which is impressive, to say the least. It’s an ingenious design with only one minor pitfall — the top and bottom bezels of the interior screen happen to be slightly thicker to accommodate the internals. While tech “nerds” (Ice Universe…) rabble about how the bezel is just 2 microns too thick hindering the screen-to-body ratio, I think this is a relatively minor deal and should just be disregarded. The phone looks fine — it’s modern, clean-looking, and sports a clever design that I value more than having slightly thinner bezels.
A healthy amount of skepticism is warranted, though — the Pixel Fold is not a cheap device — it doesn’t even come close. It’s $1,800 in the US, which is, frankly, an insane amount of money for a smartphone. This is very clearly intended to be a device for enthusiasts, not “normal people,” which was kind of already implied since this phone is made by Google, but regardless, it’s an important detail. I’m more excited for when Google inevitably subsidizes R&D/production costs and can make the phone cheaper, say $1,400-$1,500. You’re paying for a chance at buying something from the future, but even so, it’s a tough sell on Google’s part. This phone has nothing special going for it — no under-display camera, crazy charging, or incredible display. It’s an average-on-paper folding phone made by Google. And it’s that last part, the Made By Google part that I have the most faith in — I want to see what a Google-ified version of Android running on a foldable phone will look like. I have faith that Google will deliver here. For now, ignoring the price, the Pixel Fold seems like a winner. I’m really, really excited and glad that Google came out with this device — it’s about time that Samsung had competition in the US market. If there’s a company that can successfully compete with Samsung, it’s Google. The Pixel Fold will start shipping next month, and you can even get a free Pixel Watch with your purchase on the Google Store, which seems like a fair deal until the generation 2 watch comes out — I really think they’re just doing that to ditch some stock to make room for the new version, regardless. I’m excited if you can’t tell. Bring me all the foldable!
That brings me to the Pixel 7a and Pixel Tablet, 2 devices that took the backseat at the presentation on the 10th. And it’s understandable why — Google wants to show off the flashy stuff, and neither of these devices are particularly flashy. The Pixel 7a is a modest, midrange smartphone with dumbed-down Pixel 7 hardware at a new, slightly higher $500 price. That price is intriguing: what has Google done to warrant a $50 price increase over last year’s model? Tons of things, apparently — the 7a now has the Tensor G2 which provides for a major speed boost, a display that refreshes at 90hz instead of 60hz (though, as MKBHD noted in his excellent review of the handset, it doesn’t actually come shipped with 90hz enabled), better battery life, wireless charging, and a huge rear camera improvement, going from 12 MP to 64 MP. All of these changes provide a significantly better value than the Pixel 6a — that’s a lot of phone you’re getting for $500. I think the Pixel 7a is the best midrange Android phone on the market by far, thoroughly beating the Nothing Phone 1 which came close to being my favorite last year. Also, the 7a comes in a really nice coral color, which just barely makes up for the omission of a glass back this year. I’m no fan of plastic (I refuse to use the word “glasstic” to describe any material) backs on phones, but I’m willing to excuse Google this time because of how good this phone performs for the money. That being said, I wouldn’t recommend you actively go out and purchase this phone right away — the Pixel 7 has a slightly larger display and slightly better cameras for only an extra $100. I’ve always believed believe the Pixel 7 has always been “The Pixel” to get — this doesn’t change that opinion. The new Pixel 7a ships immediately, and if you’re buying one, please buy it in the coral.
The Pixel Tablet was first announced at I/O last year, and well, my initial reaction to the device was to mock Google for announcing it literally a year early even though all they had to show us was renders. But, the move made sense — Google can’t keep anything a secret nowadays (or ever, if we’re being honest). Unveil it before the press does, I guess. The tablet’s headlining feature is one that I praised last year as being cool — it can convert into what is essentially a Nest Hub Max by docking to a magnetic stand/base thing that you keep in your home. This is something I want Apple to do — it’s a really clever way of making a tablet useful when not actively in use — something that Google is notoriously really good at (ambient computing). I don’t really have much more to say about the tablet — it has an 11-inch display, Tensor G2, and 8 GB of RAM, which is to say, it has respectable specs. And the price is respectable, too: it comes in at $499. This seems like a reasonably-competent iPad competitor, but I still don’t see why Google doesn’t just resurrect the Nexus line of tablets to actually compete with the iPad Pro. I’m indifferent to this product — it has a neat docking feature, but I don’t see why you wouldn’t just buy a budget iPad if you wanted a good tablet computing experience. It’s a Google-ified tablet, what do you expect?
This I/O was different — there wasn’t much talk about Android or other mainline software enhancements. It seems like they mostly covered all of that at CES 2023. Yet, it was a packed conference — Google really did put a lot out there last week. I just can’t wait to see what Apple has cooking for WWDC. Tech season is coming, folks. 💣
