Google's True Moonshot
stratechery.com
stratechery.com
But right now they really really seem to be failing at it. Maybe the core tech in gemini is great. But it's nowhere.
Just as an anecdote, I tried to actually pay google for an AI product they claimed to have launched - the image generation, Imagen 2. And apparently, I can't. Even after tens of e-mails and a call with an account manager, the response is "uh, follow us on twitter and explain why you are good at building AI tools". Jeez, buying a service is not supposed to be like a job interview. It's supposed to be like buying the same service from OpenAI - enter credit card details and go.
So, the issue with google is that they took the wrong approach - build it in-house at a big company. What a big company has, are lawyers. Very good ones. The job of lawyers is to avoid risk. And they are great at it. However, building these sort of cutting edge services requires taking risk. And you can't really do that at a large company.
This is why Microsoft is winning - they realized that investing into a startup that has no lawyers and is willing to take risk is the right path to quickly getting to the result. This is also why dalle3 and chatgpt4 are available for everyone today. And Geimini ultra isn't.
That would be the one that MS invested in.
Well, for now at least. At some point one of those free models will best it.
I agree, there are some minions whose valuations have skyrocketed because they happen to also try and compete in this very bubble-y market, but other than that I haven't heard any "those guys' LLM product is great!" comments related to anything other than whatever OpenAI or Meta have made public.
Yes. Perhaps. That said "the model currently worth paying for" isn't the only possible business model or approach to "winning AI." It was a pretty incidental business model that unexpectedly got big.
I'm not convinced this is the final state.
AI models need users. There's having the most users. But, there's also having the right users. This isn't a linear race. Image generation. Language generation. Those work now. They'll get better... but it's still not clear what better enables. Best way to find out is by finding out what current models enable.
IMO, there's a lot more room here for bravery and creativity than beating OpenAI at whatever the race appear to be at this particular moment. The budget necessary to build applications is nothing compared to the scale of strategic investment being made.
Make the LLM chat/email client. Make the LLM project management software. CRM. tech support. All the cheap, obvious stuff.
Once upon a time, Google had a let 1000 flowers bloom approach. They did succeed in launching lots of pretty good products. They did not succeed in turning that tactical success into a strategic business success. That said, the tactic worked.
Why take the risk/effort/etc of winning the race to cloud services? Google have already lost one cloud race to google and amazon. Just build the apps yourselves. Find a framework. Do it. Don't want to run them forever? Then sell them off. Shut them down and leave the space open for startups.
Why futz around?
Not when AI agents start doing their own thing, synthesizing their own data to go beyond what humans have conversed. The main thing keeping this from happening is data corruption. I don't know how that will be solved.
Spoken like someone who has never used bing chat. Its so terrible its funny.
Google has the ability to build something better than OpenAI but that's not even a certainty
I think it's still an open field. Ability to make AIs into products and businesses... that's a lot of the game.
I agree about housing it all inside Google. Not necessarily because lawyers, or even because risk. There are advantages to both in house and outhouse approaches but... Google just isn't that good an in-house.
We do not know, at this point, what are the big businesses opportunities presented by recent AI advances. There's a lot of focus on near-future breakthroughs. IMO, many of the breakthrough products may already be possible to build now.
There's also an innovator's dilemma with the Google search mothership. One obvious area for LLM's is search replacement. "How does adwords benefit?" is not necessarily a question you want to have to answer, if your goal is search-replacement.
Then Google bought Youtube.
In the AI space it's unclear what/who they could buy -- OpenAI being obviously out of reach -- but it's possible they could find a good match.
you're all just telling each other exactly this in hundreds of different ways.
I think Google invested like $2B+ in Anthropic
On the other hand OpenAI has been quick to take money. Mistral too, got an account immediately and able to add billing details and they don’t even have a usage page so I have no idea how much I have burned through.
IIUC their open Mixtral 8x7 maps to the mistral-small model (which while impressive is still behind gpt-4).
I guess they want to keep their best model behind the API for now.
Anyway, I second that Mistral does stand a good chance to compete with openai
They bought the most promising AI startup in the world for $650m and failed to integrate it or make use of it until ChatGPT was released.
That was DeepMind, nine years ago.
Edit: I thought about the topic more while wrapping up my work work. I am on the periphery of AI at one of the large US tech companies and we've placed AI bets along many of our existing products. Every day I run into this – "I manage xyz product, we plan to add AI to help with XYZ in 2024" or we added this chat functionality for "manual job to be done." I don't claim to have insight into the future on which of these solutions will be successful for their intended clients. The pattern I see is that AI can be quickly (relative) integrated or coupled onto existing software services. Is that the secret to AI? It will permeate through our digital lives either in small or big ways – but critically it isn't one AI to rule them all. AI micro services will act like intermediaries between humans and some end service.
It is like chocolate. Why not pair chocolate with [enter any food stuff]? You could hit a home run like chocolate with peanut butter or chocolate chip cookies. Now we have chocolate everywhere including drinks, but chocolate isn't required for a tasty result. And importantly chocolate isn't always a standalone dish – it can be though.
I disagree strongly with this, and it has everything to do with how we conceptualize AI.
I get in my car and plug in my iPhone. CarPlay immediately causes the Maps app to pop up and route me to my next meeting. I can say "hey siri, set a reminder to call Mr. Jones at 3:00" and she will gladly comply. If my buddy texts me while I'm driving to the meeting and asks if I'm free for golf tomorrow, she will automatically try to pin that on my calendar. I can throw out lots of examples here, but you get the idea.
Now granted, voice recognition in Siri has been pretty bad. She struggles with a lot of basic things, like putting on the music I request. But, there's no question in my mind that these augmented reality moments are where AI is actually making a difference in our lives and represent the actual business opportunity bridgeheads. In Apple's case, they not only already control the hardware (the phone, the watch, the earbuds, the tablet) but they've also figured out how to start bridging this into other hardware like a vehicle.
• Maps doesn't understand that I don't own a car, defaults to driving sometimes
• Autocorrupt rather than autocorrect
• Calendar suggestions only work for the simplest of dates, so it suggested an event for the wrong month
• One case where it seemed to think the only timezone in the world was California
(It's not all negatives: for me, Apple has the least wrong voice transcription AI, and I do like their computational photography and definitely the ability to select text in images and Safari's website translation — but even then I don't think they're way ahead of the rest with these things, and website translation was definitely behind).
* Microsoft to do it for business users
* Google pixie does it first for personal use. Android phones might even momentarily be more desirable than iphone
* Apple eventually get there, for their walled garden
* Amazon etc fall totally behind
* And, years later, users of google docs are still waiting for basic AI help writing docs whilst google completely doesn't try in that space
AI would be much better at auto-completing whole sentences and paragraphs and suggesting rephrasing etc in docs and presentations, than at answering questions about the world.
But the race seems to be to answer questions about the world based on a 2021 dump of the internet, so ... :)
A good on-device assistant would be super useful: - instead of searching for a setting, you just tell it to change a device setting - decent appointment/reminder setting ("remind me to invite joe to my next birthday party" -> and it knows who Joe is, and how to set up such a reminder) - all kinds of search ("what was that book about startups that someone recommended to me a year ago?", "open up a tracking page for that thing I ordered last week", "show me a photo of XX I took last summer") - managing e-mails the way old-school secretaries did - "any important messages?", or "reply to everyone I'm out of office, unless it's related to X"
I make a point to not share my search history with anyobe, or any other data as much as possible. And I wont start doing so to enable some AI gadget I am more than happy to live without.
"Pixie, reply to that email for me"
"Pixie, book that car service for me and remind them they promised a free brake fluid replacement"
"Pixie order all the ingredients I need to make my wife her favourite dish"
"Pixel, buy everyone in my family birthday presents"
I will 100% pay for that capability, and as someone that uses GPT4 to basically do my job for me, it really does not seem that far off. And I do think Google has an advantage here over Apple or anyone else, not just in Hardware, but in the enormous amount of data and information they have about both people in general, and specifically me.
By the way, if I would use ChatGPT for my job, let alone letting it do for me, I'd be fired, worst case go to jail. And even if not, what would prevent my employer from firing me anyway if a basically free to use web service can do it at the same level I can?
My grandma said, "I won't send any email through a computer, I prefer to write it by hand and have it delivered by a human being".
Although maybe it will be even weirder with AI assistants sending birthday greetings to other people's AI assistants and then your AI assistant summarizing who sent you birthday greetings.
The fact is the next generation moves on and uses new tools. I don't think AI will replace our jobs but people who are good at their jobs can, with the help of AI, outcompete people who are just good at their jobs but don't use AI.
The parent you were replying to may currently be able to automate what they are doing with the help of AI but I don't think that will last long as jobs end up requiring a mix of human and AI capabilities.
But reading? On-premise? You already allow basic algorithms to read your mail, for spam control and whenever you search through your messages. Having a better algorithm (as long as it's still on device) has no difference in terms of safety.
Phone, remove the blemish from this picture.
Phone, make my hair slightly more red, and my waist thinner by one inch.
Phone, tell me who that guy in the background of my pictures is.
Phone, what time do I need to leave to get coffee on my way to work.
Those sorts of things.
Most of the world is already there, and it's not momentary. iPhones only dominate the market in North America - Android has the majority basically everywhere else
As for docs, I can see them pushing for more Gemini integration depending on how M$ Copilot goes when it reaches saturation
> Most of the world is already there, and it's not momentary. iPhones only dominate the market in North America - Android has the majority basically everywhere else
Market share is not necessarily proportional to desirability. Cars make a useful analogy here. I think a Ferrari is more desirable than a Volkswagen, but Volkswagen’s market share is much higher.
Just look at anyone under the age of 25 (35 maybe?) They can easily have long, meandering conversations with actual humans using voice, yet I see them go for text 9 times out of ten. As someone on the backside of middle age, I often find it pretty baffling. I like the succinctness of text when I need to send a quick update or ask a short question, but I normally always call someone for an in-depth conversation. But I'll see my nieces text back and forth with friends for literally hours, sometimes getting emotionally worked up, and I'm thinking "OMG, just pick up the phone to your ear and just talk to them."
But I think the reason people prefer texting is the same reason most people still prefer typing, despite tech that, these days, could easily transcribe with great accuracy. At least for me, typing frees up my brain to actually move faster. When typing, I can think about the next phrase or sentence. When talking, I find it much more difficult to "think ahead", so to speak.
So I'm really skeptical that voice interfaces will be "the wave of the future". Sure, I use OK Google a lot, but basically for the same sets of commands as everyone else (What's the weather? Set my alarm. What's next on my calendar? Etc. etc.) Occasionally I'll ask it "search-like" questions. Perhaps I suffer from a dearth of imagination, but I just have a hard time believing long voice conversations with a machine are something most folks would want.
For me, I prefer to dictate to my Android phone in contexts where most people choose to type, such as in a short message. It's faster than typing on-screen, and the way I think I am able to compose ahead in a way I can't do while typing. The dictation is so good these days that there are relatively few mistakes and ambiguities to correct, and the UIs for doing so have become easier.
I wonder if this would change as we got more familiar with interacting with a voice AI. I think a lot of the extra brain power that goes into talking versus typing is the assumed need to keep talking at a constant pace because that is what a human listener expects. If we were more comfortable with pausing while talking and not feeling like we need to always know what to say next as soon as we get to the end of a word, it might not take as much brain power to speak to a computer anymore.
Can't be worse than a bus full of people recording voice messages aloud and listening to them on loudspeaker which is daily life in say China. I would be lying if I said it's not maddening
No such issue with robots.
Where do you live that they prefer texting?
My acquaintances in that age range love sending voice (and fucking video, yes yes) messages. And I mean not intimate chats with close friends or relatives where you want to hear the voice and see the face. No this is just how they convey info. Including some people from UK.
This is all just speculation, I'm not really a texter, but I do find it interesting when limitations might turn out to be features.
An AI assistant fundamentally hits at Google's gut: Search ad revenue. Yeah, Google can make dozens of AI demos. Are they truly ready to put Gemini as the default interaction model, pitting it against their search? Can they give super accurate answers to questions, without the potential of 3 sponsored results on top, in such a way that it would make their search obsolete?
Shareholders "wish" to see competitive demos, to not be left behind, etc. But are shareholders & market ready for a 10% balance sheet revenue drop that comes with making an AI assistant a go-to product instead of search?
AI in Google could end up getting crippled, not by design, but by the environment and "thou shalt not touch search revenue" constraints under which it operates in.
It would be interesting if Alphabet (and Alphabet's moonshots) totally distances itself from Google instead of being an internal structure sharing the same tickers, and treat Gemini as an all-out cannibalistic competitor to Google.
Probably not, since they have not figured out, how to integrate the ads into the AI.
An additional problem they have is that running AI at scale is super expensive. Even for Google. Add to that legal risks, pressure from regulators in different countries, privacy concerns, copyright issues, etc. and you get a whole lot of risk and friction. Add to that the usual organizational infighting and politics and you pretty much have a company incapable of doing anything.
Yes. Long term, if you control peoples attention and their flow of information, you will almost certainly figure out a way to control their pocketbooks.
I understand this argument, having made it multiple times in the past - any development at google that threatens the cash cow will, ultimately, go nowhere. Why, after all, would they spend money developing something that reduces their income.
After a little bit of thinking about it, I offer an alternate future for google+AI: the AI improves their ad relevancy to such a degree that it makes google search more valuable, not less.
With all the data they have on each individual using their search, it is not inconceivable that that data + an inferior model beats out a superior model that has less context.
More valuable to who? In this case it seems like more relevant ads increases the value of Google Search to advertisers but not necessarily users.
If, as a user, I see ads that are more directly targeted at me, that is a negative value in a search engine. I never want targeted ads. I only want to see what I searched for.
That said, for the small minority of users like me, Google may never be able to overcome their “advertising company” label. And that label is an instant black list.
> More valuable to who?
To the paying customers - advertisers.
> In this case it seems like more relevant ads increases the value of Google Search to advertisers but not necessarily users.
So? Google doesn't care about its users, and hasn't for a very long time now.
Google became profitable purely because they delivered the most value to advertisers, by way of having the most eyeballs. More relevant advertisements are not going to reduce the number of eyeballs that google sells.
> If, as a user, I see ads that are more directly targeted at me, that is a negative value in a search engine.
And those users will be lost in the noise, a statistical rounding error, if that. Losing unprofitable users helps Google become even more profitable.
Google as a products and services company - Google with Apple's business model - is something I would've been a happy customer of for the last decade or two easily. Google as an ads company is an entity I go to great efforts to remove from my life.
I am huge fan of youtube and the incredible content and creators on the platform. If it was not owned by Google, I would probably happily pay 50 bucks a month for an ad-free and tracking-free experience.
But instead I've mostly left the platform. I don't want the company with eyes all over the internet to also know exactly what I watch and when. It's creepy to have a single company know exactly what you did/read/watched for most of your waking hours.
That's why I personally would never pay Google other than for very basic and needed stuff like GMail, I wouldn't want my email account nuked and my life momentarily turned upside down because of some payment misunderstandings related to their streaming service, let's say.
That's not the real question though - the real question is whether people find iOS compelling enough to pay a premium for using it and put up with the ecosystem lock-in. And the user experience is only superior when you're used to it, as a longtime Android user I regularly get frustrated when having to use an iOS device...
It’s only paying a premium at the lowest end of the market - if you’re buying a Pixel, you’re paying as much or even more than someone buying an equivalent iPhone. The ecosystem switching cost is real but given how most Android users I know complain about app developers favoring iOS, I’m not sure how substantial that is.
This is bad for Google as google’s main feature is search. Ai assistants will replace search and kill the main revenue stream which is ad clicks.
Google’s secret plan is to eventually canabilize it’s pixel phone with an ‘agent first’ device to try to beat Apple, MS & OpenAI with a horizontal offering that is significantly better than the fragmented world of ChatGPT &iOS & azure.
But it’s worth remembering and I think the article fails to point this out. Agents will still ‘recommend’ things, you ask them to book you a flight - they still have to recommend a couple of options out of many - the agents need to decide and ultimately that decision is the same as choosing who to place at the top of the search results page - whoever wins the agent wars will win a significant proportion of Google search revenue as referral fees AND a significant proportion of IOS App Store revenue.
It really is winner takes all.
Now imagine you get some very bad data loss bug affecting just your company account. What chances are there you will ever have your problem looked at by Google? I totally believe it's 0% unless you get to HN frontpage.
It's not like Dropbox have best support ever, but I totally believe that getting a living human to look at your problem would be so much easier with them. With Google it's just impossible.
Paradoxically we are in a phase of technological stagnation. Cloud movement was/is maybe the latest paradigm shift for a couple of years to decades to come.
AI will accelerate features and won’t mark a product category itself, same as gifted people are capable of outperforming others in intellectual fields, but don’t necessarily need to.
The mundane stuff like generating text is what is AI paradoxically needed for. (Imagine that, that humanity’s greatest gift is now outsourced.)
This is what makes AI so hard to grasp. We know what it is, but it is hard to applying it in concrete business contexts. “OK Google analyze my company with 100.000 workers” won’t happen soon. Society needs to take care of possible side effects first.
On the other side, there will be a ton of cowards who will say the computer told them to do X, where X is the thing they wanted to do all along. And the neat part is that you don't even need to ask a computer for that, you just pretend you did. And in the end if Xbturns out to be a shit choice, you can use the famous software error-excuse and be off the hook immediately, because apparently liability ends there automatically or something.
They will make something remotely nice, half ass the final 20% of it, leave it to rot, complain that there have been only X billion dollars of profits and not Y, and kill it off.
How they are not sunsetting Search yet, I cannot fathom.
This is Google / Alphabet's biggest problem. Unlike Microsoft that has Enterprise, Gaming, Cloud and multiple other 10B+ businesses that don't rely on ads, Google has only cloud.
And Google Cloud is a distant third where they actively sabotage their own success by not caring about customers. 90% of Google revenue comes from ads. From Search, youtube, Adsense e.t.c
If they build an AI that gives exactly what the user wants without ads, Google's stakeholders would fire the CEO the next day.
My perception is that Google is in AI race to be relevant and capture the talent, but they don't want to actively put AI anywhere near their products that would eat into their ad revenue.
Google Brain/Deepmind produces great research, they have the most number of published papers, by a big margin, but Google is not the place if you want to work on productionizing AI.
Top AI Researchers probably make the most money being at the big tech labs - DeepMind / Open AI / Meta FAIR. I've heard comp being into the 10s of millions per year.
How does he square any of that with the rest of the preceding article? Google having a technical advantage on paper and still fumbling the ball is their M.O. for a decade. Their ecosystem of devices, even just sticking to the Google-branded hardware, are inconsistent crap. Take a simple, first part app like Google Calendar and look at how sloppily and poorly it works across Pixel phones, Android Wear devices (including the Pixel Watch) and Home devices. Google sucks at building polished, consumer products.
Google is by far the best company positioned for AI assistants? Is this guy forgetting about Apple, the company that has a much stronger hardware ecosystem and that will demonstrate this in the coming months with a huge hardware launch that only they could do, leveraging a huge base of iOS apps, AirPod users, etc. Apple is building a network of personal hardware devices with a weak but well-integrated personal assistant. In 1-2 years, Siri will get a massive improvement based on recent advances in LLMs and it will be 80% as good as Google’s then probably thrice rebranded AI, but it will be 300% more polished and install automatically on hardware people actually like to use.
Uh, I think you're living in another universe. I have all of those products and I have no issues with the Calendar experience. What use case do you think works poorly exactly?
Also that idea Google will build such a magically superior AI to every else is a bold claim. Can it make a competitive one? Yes. Can it make one so good that it makes every other AI obselete and everyone trade their iPhone for Pixel? I think that's unlikely given the level of competition
Unfortunately, ball-less wonder milquetoast Sundar has none of the above attributes.
He was hired 10 years ago not to rock the boat, but to stay the course, a strategy that requires a CEO with all the exact opposite attributes: meek, boring, visionless and all over the place.
If textbook quality data is needed, then we are basically limited by the current best LLM’s ability to create synthetic textbooks.
Or perhaps this is a path Microsoft is trying (and presumably openai) due to a lack of good non-synthetic data.
> Google’s collection of moonshots — from Waymo to Google Fiber to Nest to Project Wing to Verily to Project Loon (and the list goes on) — have mostly been science projects ....
...a car service rather far afield from Google’s mission statement “to organize the world’s information
...What if “I’m Feeling Lucky” were not a whimsical button on a spartan home page, but the default way of interacting with all of the world’s information? What if an AI Assistant were so good, and so natural
So... I think we should distinguish between "moonshot" and "silver bullet." One is a big, difficult goal that can be a approached with lots of determination, resources and such. The other is a future breakthrough that just fixes everything.
Google has always struggled making (great) technology and concepts into products, and products into businesses. The biggest miss, IMO, was cloud. Msft & Amzn relative successes highlight where google isn't strategically strong.
Waymo might be the biggest investment (probably >$100bn risked). Cloud is the large business category that actually exists. Google should have been here, considering where everyone was circa 2008. Google had the tech, the concepts, even the products. It wasn't effective at making that a great business.
The "OK Google" assistant story tell objectively, because no one else has done a great job with voice interfaces either. That said I think it demonstrate the difficulty of going "concept to products."
IMO, the problem with voice assistant has been a problem of imagination. Voice is a UI paradigm. What are the key use cases, where this new UI paradigm is powerful? They never invented it.
Anyway... I think the strategic logic is flawed... if that is indeed the strategic logic It's "singularity thinking." An expectation that version N+2 makes version N=1 obsolete. He who attains the GPTn, owns driving, personal computing, etc.
>The potential payoff, though, is astronomical: a world with Pixie everywhere means a world where Google makes real money from selling hardware, in addition to services for enterprises and schools, and cloud services
So this is what I mean. A "moonshot" would be defining these and going after them with real big intent. Not one that considers everything side effects of some big breakthrough that makes all linear approaches irrelevant. Voice UIs, even self driving, whatever wing's mission is.... these aren't impossible ideas even with current science & computing power. They're just hard. Requiring imagination. Risk. Vision. Strategy.
Drones are a method. Delivery is the task. Human-like drivers are a method. Transport is the task. Voice recognition. LLMs. These are methods. Not tasks. If you're doing silver bullet, it's nice to avoid defining the task. If you're doing moonshots, you want to be brutalist in defining the task.
> search
many people disagree..
this is about how many people now going to ChatGPT while were going to Google year ago for similar searches
The potential payoff, though, is astronomical: a world with Pixie everywhere means a world where Google makes real money from selling hardware, in addition to services for enterprises and schools, and cloud services that leverage Google’s infrastructure to provide the same capabilities to businesses. Moreover, it’s a world where Google is truly integrated: the company already makes the chips, in both its phones and its data centers, it makes the models, and it does it all with the largest collection of data in the world.
Yet another new Google project is simply met with “ok, but when will it be canceled”.
You currently still need search too, for AI's to access data newer than their training set, but I wonder how that's going to evolve over time ... Could see AI's being updated increasingly frequently and tapping direct into news sources (not via search) for news/sports/etc.
But this distinctly reminds me about articles around about the time of the first iPhone release which posited that Microsoft was in a unique position to ultimately win the smartphone market.
After all it was dominant on the desktop and Windows Mobile already had a decent (albeit not dominant) marketshare in smartphones at the time.
They have been "uniquely positioned" as you put it to win the AI race for the past 15 years now.
And they have done exactly zilch (other than produce an OpenAI-wannabe fake video with Gemini a couple of weeks back).
I see no reason why that will change in any way.
Astronomical payoffs is not something Google is culturally capable of anymore, because they require risk taking, and strictly no one has any incentive to take any kind of risks at Google anymore.