Google Invests Almost $400M in ChatGPT Rival Anthropic
bloomberg.com
bloomberg.com
Could they acquire an OpenAI competitor, maybe, but anti-trust is a pretty big problem right now. It's most likely that we're just going the route of everyone copying the form factor and whoever has the best experience/data and speed of response wins.
I agree with some of your points regarding scalability and add the fact that Transformer architecture (which GPT-3 was trained on) was created at Google. They also have all the data in the world to train a better model.
Interesting to see what Google will come up with.
And how many of those new products cannibalized their current cash cow?
Let's not act like ChatGPT is not a completely different sort of threat than game streaming on Stadia or some dumb product like that.
In Google and ChatGPT's case, a chat bot that functions like an answering machine needs almost exactly the same kind of people, infrastructure and workflows as their current search team. Google could at anytime decide to make those work on a ChatGPT clone, and that would be supported by all of their search pipelines and data models and release workflows they designed for search.
Will chat make less money than search? Probably. But search isn't such a large part of Google, worst case Google will have to kill some more side projects.
Chat isn't designed for project management, but your average slack user thinks that if everyone vents their opinions that somehow a consensus will develop - but more often than not, nothing gets done.
MS Teams won because it brings collaborative editing to Office. It’s ressource hungry and annoying but a lot more useful than Slack ever was.
HN people see politics behind everything these days, which is sad, because in this instance the business reasons are so obvious that it makes these comments sound like 4chan rants.
ChatGPT is slowing down at its current scale. The scale on day one for Google or Apple will be elephantine compared to the scale at ChatGPT right now. Someone has to engineer a credible solution to those issues.
Then there's legal.
And don't even get me started on UX and monetization. For Google, there has to be some way to monetize this before you butcher your market. And believe me, Google has a lot of very smart people searching for a monetization model. It really is just not as easy as people are making it out to be.
Ex:
"Sure, I can answer that question! [But first, were you aware that 9 out of 10 americans surveyed prefer Fruit of the Loom brand underwear? We weren't! That's why we sent our designers undercover in public restrooms to yada yada yada for 4 more sentences. Click here to learn more!]
Now, on the question of sigmoid vs relu, [Remember to watch Last of Us! Now streaming on HBO Max! blah blah blah] in most back propaga....
<User opens new tab to ChatGPT or Siri+>
> of very smart people searching
> for a monetization model
Nope. The entire hiring system was not to hire people like that. They need some “street smarts” / “cowboys” - but these people did not pass Google interviews. Or they were not able to prosper at Google.
Or even if such people are hired, they aren’t empowered to the extent that they can impact the company.
It’s an engineers playground and it shows.
I don't think expect Google to provide something better than openai, they will likely provide something just because everyone else is doing it, but I think openai will keep a technological edge, purely because of the people they have
But this is the point. The technically superior product doesn’t necessarily get the market share due to factors like the Teams distribution model.
It’s not hard to imagine similar strategies from the existing big players.
Giving it away as a part of something many already pay for let an inferior product like Teams gain marketshare.
The problem in this case with Google, is their main monetization method doesn't fit nicely with a ChatGPT like interface.
And it's not that hard to imagine at all: ChatGPT as it is shows how much coercing out and baking in certain things into a model can create a pretty serious product--the whole thing is honestly begging for ad injections.
In general we live very far away from a world where these things aren't immediately going to be bastardized for profit, however well it "works" to do that. Even if this was all truly revolutionary, it wouldn't be worth it if it doesn't extract money from consumers, it wouldn't even count as "revolutionary" in the way we use that word these days. There is no hope for a non-adsensed ChatGPT, thinking otherwise is like wondering if the sun will rise tomorrow. Maybe it wont be OpenAI, or Google, but whatever it is, it will rise to the top.
In general, there will never be a technology in our world that can achieve much through solely its own merits, it can only ever succeed by virtue of how efficiently it can turn around money. We are in a small honeymoon right now, but don't forget the rules!
That is certainly what happened for Google Plus – who even remembers Facebook now?
Google hasn’t shipped a big success since the 2000s, with G+ and the Reader shutdown marking the end of the period when they built things which were good for users rather than what their executives wanted to sell ads. It’s possible that this could be the big reversal but given how much worse search has become I’m skeptical that they can do so without different managers at ggd top.
Google Plus launched in 2011.
I don’t think it’s just a coincidence that 2008 is also when they bought Double Click.
So yeah - what have they released after that?
Stadia had potential, but google killed it. What else?
seriously though they have just been building their base while also trying to grow. they do have a bit of a short attention span regarding growing though. if you don't catch on like gmail yer gone.
It took a generation for xbox to "catch on". It really took off only in x360 era and later. Imagine if the xbox was killed off in 2004.
Stadia was shut down after 3 years and first party devs were laid off even earlier. That's completely insane - it takes years to develop proper AAA games.
Just look what SONY first party studios are doing. Imagine if Naughty Dog was dissolved after 3 years.
The Wii U was weak, but Nintendo didn't leave gamedev business. They learned from their mistakes and the Switch became a huge success.
PSNow streaming service is still running just fine and SONY doesn't plan to shut it down. So game streaming business model is certainly viable.
It seems like that Google didn't even try to make Stadia a success.
I use Apple and iOS devices because I have to do iOS development for work… I use Linux and Windows where they make sense and I live a cross platform lifestyle… but I think I’ve actively used a chromecast once in my entire life, because it’s never been better than plugging my own shit into a hdmi port with a cable from my bag…
Whats the point beyond TVs that aren’t smart? (Mine isn’t smart… but I’ve got an Apple TV as a dev device for testing.)
Is there a point beyond that? I’m genuinely curious as someone that’s given away literally a dozen chromecast dongles because I give zero fucks about the Google ecosystem of lock in garbage that might be abandoned tomorrow… oh and fuck attaching physical spyware dongles to my home network… cannot emphasise that enough.
However it is not a superior product to Slack. It has a lot of basic issues.
However Google could probably do better. My hope is they add it to G cloud.
I want to have a simpler way to deploy infra without needing to fuck with yaml and know a million AWS specific things. Gcloud is already a decent experience compared to AWS, and maybe it can be even better if spinning up instances can be made so easy that a product manager can just ask to have a build ready to have feature demo for a customer without needing engineers to do anything prior . Let them focus on building and have Gcloud robust enough to handle it well.
With the "worse" product
Would they have obliterated it if Microsoft wasn't already in bed with all of these corporations?
Why didn't they ship something like ChatGPT earlier? Lack of product vision? Perfectionism?
Is this Xerox vs Apple all over again ?
Tbf ChatGPT was far from production quality for serious applications, lots of misinformation and you can make it produce very offensive content. It is a good for toying around but you cannot take the output seriously.
This is why "jailbreaking" is a thing. Once you convince the model that it's OK, it'll let you do anything from then on.
-Emily
I think it's right that the owners understand what the weaponization of ChatGPT could do and prevent it, and I think we need laws (and fast) before weaponized AI like ChatGPT turns into a disaster for humanity
I welcome idiot savants.
They suffer from "scared to release an imperfect product that damages PR"-itis.
They are so scared that an AI with Google branding might say something inappropriate, that they prefer to release nothing at all.
For context, they put a lot of engineering effort into making sure "Okay Google" for the voice assistant couldn't be modded to "Yo Dick" or something else off-brand, simply because they were worried about media backlash.
Nobody knows FB owns Instagram.
Nobody is going to care if Google releases a ChatGPT bot on a different website.
It would be a 'you can use all google internal code and infrastructure, for free, as long as you don't use google branding and don't use our user data' deal.
The companies would have the benefits of google tech, without being tied down by internal politics, procedures, risk management, etc.
They could afford to do things like just release a product to the world rather than have a legal and PR team go over every detail for every law in every country worldwide as part of a 6 month long review process.
The closest Google had to it was Area 120, which has also been abandoned?
I might bootstrap revenue without much hiring, so that business success means the team does better financially than the sure-thing FIRE if they'd just taken a FAANG job and grinded promotions.
This business giving away 80% equity would lean on the value of IP, and on knowing how to be very effective with solo/small teams -- not 'scale' in the sense of trying to herd hundreds of software and product and business people. (This is another thing that big orgs can have trouble doing, and so sometimes do insulated spinoffs/units instead, just to unburden people.)
[0] https://area120.google.com/
[1] https://techcrunch.com/2021/10/05/technical-interview-platfo...
They're pretty upfront with it... It says "By Meta" on the launch screen
That might have been true at some point - but a combination of their extremely loud rebranding of both to Meta, and the media constantly saying 'Instagram, which is owned by FB' or words to that effect has changed it. Even my elderly parents are aware Instagram is owned by FB / Meta.
Even my non tech friends know.
More things that don't work:
1. Delete the last screenshot.
2. Set timezone to est
3. Turn on hotspot
4. Turn on nearby sharw
However, it should at least just ask for permission the first time you try to do that.
Finance and management, knowing their bread and butter is as advertising distribution company, undermine, kill, and dilute any projects that are not obviously, undeniably directly tied to improving their advertising distribution revenues. Their big problem is, they never communicated this core identity to the larger corporation, and somehow search is allowed to coast...
Neither of these two things are true. It's 51% owned by the founders and public companies are not required to only care about profit.
Public companies are legally required to put shareholder interest first - that’s what the whole point of this facet of capitalism is: to provide profit to the “owner” class of society, at the cost of worker exploitation.
Get Adam Smith’s boot out of your mouth and recognize that literally anything a corporation does is for the goal of profit. Corporations are not human - and only have a decision making process that puts “return” first. Zero of them are “not evil.”
No, they are not. And they definitely aren't required to care about the interests of 49% of them.
> Get Adam Smith’s boot out of your mouth and recognize that literally anything a corporation does is for the goal of profit.
Is everything an animal does for the purpose of finding food? (No.)
"At Google, Character’s co-founders Shazeer and Daniel De Freitas were instrumental in developing Google’s LaMDA language model, used for natural language processing research. They decided to launch their own startup after getting frustrated with big company bureaucracies, Shazeer previously told The Information."
[0]: https://www.theinformation.com/articles/character-seeks-250-...
ChatGPT is very cool but they don't make money from it and GPT-3 had a paid public API for a year now and I read that the money they make from it is very little, insignificant compared to the funding they get from VCs.
Currently, there's no clear path for monetisation but has the capacity to take away a lot of Google's use cases. In essence, it's like Napster destroying the Music CD business.
Maybe it's also a bit like Kodak: Invents the technology but its current business is too good to be messed around with. This new tech is the future but we are happy in the moment.
If Google wanted - they could've easily released ChatGPT version that is 100x better than current one considering they invented the technology and they have thousands times more and better data to train it.
They just weren't interested.
I wouldn't be surprised if they used anonymized gmail data though. Even if they removed all proper nouns it'd probably provide a lot of extra tokens for training.
The impressive thing with chatGPT to me is how well it understand what you want even with very sparse input. Even if it gives wrong answers it still feels like you are both on the same page. That seems like the secret sauce even if a larger language model would give more correct output. I wouldn't be shocked at all if Google doesn't have anything currently that feels the way chatGPT does when it comes to interaction and now they are racing to build exactly that.
I don't think it's that, either, given the rate at which they are involved in overt censorship, over the last years.
I believe this tech is so different and revolutionary, that the lawyers have yet to wrap their head around it. I predict a million lawsuits incoming any time now. This I believe is the primary reason for Google not releasing any product before the water gets tested. For what its worth Google already has access to lots and lots of data that it owns, that it can train its AI on (Photos, Gmail, Docs, obviously Search and probably YouTube as well)
I love it, but it’s not ready for all Google users yet.
The simplest answer is that the people with the know-how did not work at Google. Knowing why they didn't is not so simple, though.
I've seen so many cool ideas thrown around by devs that could actually go onto be monetized by companies go wasted. To think they'd be worked on for free (spare time) and often better then outsourcing because nothing drives true innovation than passion.
Focus on BAU, keep your head down and be lucky you're not getting fired.
The fact that they are freaking out is actually a good sign though
They back stepped, and then watched someone else capture the market.
I would say they have just in different form. For a few years now when you google something you're given a quote/excerpt at the top of the results that typically answers your question.
In order to launch something you have to need to launch it. Either because you're losing market share, or because you need the revenue, or because your customers expect it. Google rarely has any of those things.
If that's the case they're not working towards a "better" model, they're probably more concerned with a cheaper one.
And maybe something where you can make sure to construct the answer using their knowledge graph instead of hoping the black box doesn't just give plausible sounding replies that are wrong.
Google makes a lot of noise but delivers little, and what they deliver is almost never the best in the world. I think maybe they made a mistake with TPUs and the PyTorch-NVIDIA path seems to be more productive and can deploy more advanced LLMs.
There was a time when I was dying for an invite to Gmail or Google Wave, but that time has long passed and nothing in-between caused that level of excitement. What I appreciate most from Google is YouTube, and not for the UI or recommendations but for simply being the largest video platform with everything you could want in music and education. If I lost access to YouTube internet would be dead for me, well, almost dead.
Google Maps comes to mind.
And of course globally its color scheme and visual hierarchy are stuck in 2005.
What gets me the most excited about systems like ChatGPT is not the systems as they are now, but rather all of the possible hybrid architectures, mixing in old tech and new ideas. I think ChatGPT is just a little first step.
- Eric Schmidt (former Google CEO/Chairman), Series A
- Sam Bankman-Fried, lead investor in Series B
- Caroline Ellison, Series B
https://www.crunchbase.com/organization/anthropic/company_fi...
In general, that’s because the the clawback period is short: 90 days before filing unless the transaction was between insiders (which no one has alleged of the Anthropic investment). In this specific case, the equity has probqbly appreciated. OpenAI is Anthropic’s closest comparable and its valuation went through the roof. Even after allowing for transaction costs, FTX trustees will probably get more for the equity than it cost.
So, the question is likely this: who will buy a ~10% stake in Anthropic?
Bing on the other hand just don't care that much and is willing to integrate it to get more traffic.
Why can't Google use the user's prompt (and AI's output) to display targeted Ads?
I'd guess given the amount of self-cannibalization Google has been doing the last few years, the fervor with which they've been sticking ads everywhere even to the clear detriment of utility suggests they're struggling to present black numbers through growth.
Having to operate at conversational AI service which is competing with their own main offering, at a loss, to run a competitor out of business is not what they want to do. Not with bearish investors, and especially not given the regulatory scrutiny they're under.
They can't even faithfully follow a search term, and instead return exactly what you wanted to avoid. Try finding "the shortest time for crossing the English Channel entirely on foot", it will tell you all about swimming, boats, even hovercrafts, ferries, but not about an immigrant trying to walk the tunnel.
When a query fails and gives bad results, it's not clear why it has failed or how to fix it. This is above all very confusing and frustrating.
In part I think the problem is the insistence on using the same search box for both modalities of search. This is straight up a poor design choice.
It's also worth mentioning that they have an inherent conflict of interest in the business model. Since they sell ads, presenting clean search results with few ads would actually hurt their bottom line. But this is a limitation in their business model and not their tech.
Google's dominance might have ended, but this is a new market. Various players will try different business models and I'm sure we'll see someone offering an ad-based product
- No 6+ ads on the search results page
- No direct traffic being diverted to publishers (which display AdSense ads)
In fairness, once the investors start putting pressure of monetization on these AI chatbots, and ads and product placement starts to creep in AI generated content[1], that would make it easier by people, teachers and evaluators, to detect it, brought to you by Carl's Jr.[2].
...Google has already "improved" its search with AI. More and more AI "improvement" each year for the past decade.
I can imagine what further Google AI "improvements" will look like as it frantically chases ChatGPT (which time might well show is a smaller search niche than plain-text search)
If I were in charge of Google, I'd dumb search down again - dumb it down to the point that it provides useful results.
I don't think you understand the extent to which search results have been broken by an arms race of SEO and spammers.
Contemporary Google search would be incredible on the internet of 2005, but it's operating in an adversarial environment (and Google in practice has to restrain itself for antitrust concerns).
And for what it's worth, I think AI generated search results will actually be more resistant to spamming, since it can more deeply understand content to determine whether is it spam or not, in the same way email spam is mostly a solved problem. Of course, if entities like OpenAI become "evil", then it is a simple matter for them to render the search less useful, in the pursuit of profits.
And I never heard anyone talk about sending people to more valuable sites. Google makes most of their money from search, not on third party ads they might monetize on whatever site you get to, if even a tiny fraction stop using search due to dissatisfactory search results then that is a net loss. So from what I know the issue is scope creep and technical depth for sure.
Also there isn't really one search team with one search algorithm, different kinds of searches has different people, so whatever you search for might have someone who isn't good at their job. But, to go with the greed reason you are looking for, I'd bet that if people start seeing bad results for search queries that are moneyizable they would quickly replace the person responsible for it and fix it. On the other hand, low value searches from people who aren't looking to buy anything probably costs more for them to run than they get back in ad revenue, I wouldn't be surprised if nobody at Google would notice if those got worse. In addition, since Google mostly uses internal tools with internal help pages and internal version of stack overflow their programmers wouldn't notice that it got worse either, and things that neither programmers nor money cares about wont get fixed.
Lastly, I quit Google years ago and have no stocks or friends left there, I don't care about them. I just wanted to clarify a few things. Also there are a lot of money grubbers at Google, but I mostly saw that in the ads organisation, search and ads are very separated though, ads doesn't have power over search since most ad money comes from search and search is the founders darling. So when things are broken its more likely to be incompetence or bloat or something like that than evil. At least for now, will change in the future as culture there normalizes to the rest of Google/rest of companies.
>The Series B follows the company raising $124 million in a Series A round in 2021. The Series B round was led by Sam Bankman-Fried, CEO of FTX. The round also included participation from Caroline Ellison, Jim McClave, Nishad Singh, Jaan Tallinn, and the Center for Emerging Risk Research (CERR).
Sometimes funds don't invest because of the product or the company, sometimes it's just to take over another investor's position (or make the other investor's position look better).
When things go really wrong, the ultimate solution is to use the company that has had a successful exit to buy a failing company.
For what it's worth, when MtGox blew up, it took years to recover and distribute what Bitcoins were left. Everyone was frozen out, but when they finally got their money back it was worth far more than when MtGox blew up. I have friends who are sitting on millions thanks to this. They probably would have sold on the way up or during one of the crashes. But thanks to the illiquidity during bankruptcy, they're far better off now.
To me it seems like Google isn't lacking internal ML talent, but more like it's lacking management, direction, vision and focus in the ML space. Buying more companies while staying fragmented and directionless won't necessarily give them a win.
Talent acquisition.
Also : Deepmind is based in the UK and does whatever the F*CK they want inside Google instead of actually contributing to Google's bottom line.
DeepMind isn't UK exclusive. Ian Goodfellow works there as well from the Bay Area, AFAIK.
>does whatever the F*CK they want inside Google instead of contributing to Google's bottom line
And that seems like the bigger problem with Google, not the lack of proper talent, which I also raised in my argument.
You can't successfully steer a ship when everyone runs amuck doing whatever they want. Making the ship larger won't fix this.
Sure, but that's what everyone does at Google. They don't hire engineers to make money, they hire them because Brin and Page enjoy keeping engineers around like pets.
Funny, but inaccurate.
First, Brin has stopped giving a damn about Google a very long time ago to go and enjoy his life.
I suspect from around the time of the "China incident" where both he and Page realized that having ethical positions at the scale Google had reached back then was just not possible anymore.
Second, Page held on for a while longer, but eventually followed Brin in not giving two f*cks any more and just enjoying their billions.
Page then proceeded to pick the most milquetoast personality in his entourage as his successor (Sundar Pichai), probably because he thought he was the candidate most likely to inflict the least amount of damage with the mighty thing he was handing over (by doing nothing with it).
No, I think the explanation is very simple: Google believes that acquiring top talent is the path to further success.
They're correct but fail to see that it is only a tiny first step on a very long road.
Namely: once you've got the talent, you must:
a. Not spoil them rotten by throwing oodles of money at them
b. Direct them towards worthy goals
c. Focus on execution by actually *managing* them
d. Not allow them to play the "hire me for 3M USD a year, I'm an AI diva" musical chair games and actually deliver tangible value.
Same reason they have both DeepMind and Google Brain. (I’m guessing)
Because Anthropic is a splinter group from OpenAI, who are Google's #1 rivals. Anthropic was founded by Dario Amodei, who was the VP of Research at OpenAI and previously led the GPT-3 project. They are buying GPT-3 rivals.
Ding! Ding! Ding! Give that man the prize, he guessed the winning answer!
Also this quote is interesting [2]
Anthropic was formed in 2021 when a group of researchers led by Dario Amodei left OpenAI after a disagreement over the company’s direction. They were concerned that Microsoft’s first investment in OpenAI would set it on a more commercial path and detract from its original focus on the safety of advanced AI.
As much as I enjoy using OpenAI’s (and Hugging Face’s) models and infrastructure, I also like to see lots of competition.
Cory Doctorow recently wrote [1] how social platforms start out by giving a lot of value to users, then short change users to give value to advertisers, and finally short change advertisers to take huge profits for themselves. I wonder if the large language model businesses will see a similar trajectory.
A brief look at the economic circumstances of the "haves" and the physical circumstances of the have-nots across the West coast under the hard labor of the SF Fed will make this abundantly clear to basically anyone.
It's a place where every cause is an unbearable burden and simultaneously the only hope in perpetuity until hush gifts and roadblocks rain down from Capital wherever.
Further, their expenditures are why you've heard that Elon Musk is a "robber baron" when in reality it is these companies that are only investing in their own funny money while the quality of their products dwindle, and often fall catastrophically as they compete and muscle out the old guard rather than cooperating and improving things. Or even better collude and calcify investment in places where growth is truly needed, like the telecoma.
Although at google scale it might make sense to eliminate competitors via acquisition
By investing in a competitor, they at least indirectly get their foot in the door of someone who releases to the public, even if the product quality is worse and the quality-bar for release is lower.
Also Google: "We need to enhance search with AI generated content"
edit: last sentence sounds kinda weird given recent layoffs round
Hasn’t there existed an entire department for 10+ years called Google Brain that should have been working on creating their own “ChatGPT”?
Is this a failure due to how Google has structured their initiatives/promotion review process?
I really hope they have tested some prototype extensively before signing the deal. Running this kind of stuff on GCP could be a massive clusterfuck.
Are you thinking they’ll let it free and train a chatGPT on GCP activity/log data or something? What would be the point of that?
---
1) How does LaMDA compare to GPT-3?
"LaMDA and GPT-3 are both language models developed by OpenAI, but they have different focuses and capabilities. [...]"
---
2) Why is it referred to as "Google LaMDA" if it is an OpenAI technology?
"It may be referred to as "Google LaMDA" due to confusion or incorrect information, as LaMDA is a technology developed and owned by OpenAI, not Google. [...]"
https://blog.google/technology/ai/lamda/amp/
Without provenance GPT is only useful for things where you don’t care about correctness, or one’s where you already know or can find the answer.
When in doubt choose fast. When not in doubt, choose fast.
Google, late to the game, will have anthropic which will be killed off/extinct within a few years anyways. The name may reference the anthropic principle but I’m going with a rewording to “Anthropic extinction”