Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
×
Google AI

Google Dusts Off the Failed Google+ Playbook To Fight ChatGPT (arstechnica.com) 52

According to Bloomberg, Google wants to build AI into everything to fight OpenAI's ChatGPT. Google issued "a directive that all of its most important products -- those with more than a billion users -- must incorporate generative AI within months."

Ars Technica's Ron Amadeo likens it to the company's failed Google+ playbook from 2011. To combat Facebook's rising popularity, then-Google CEO Larry Page directed employees to build social features into everything. YouTube comments were tied to Google+, Gmail addresses required a Google+ account, Google Search had "+1" buttons, and a "real name" policy was instituted, among other things. "That forced integration strategy was an abject failure, and after a few years of Google's social panic, all of Google+'s integrations were removed, and the service was eventually shut down," writes Amadeo. An anonymous reader shares an excerpt from Amadeo's report: We wrote last month that Google's ChatGPT panic seemed a lot like its response to Google+, and several employees relayed that same sentiment to Bloomberg. Just like with G+, the report added that "current and former employees say at least some Googlers' ratings and reviews will likely be influenced by their ability to integrate generative AI into their work."

AI is one of the few areas of Google that CEO Sundar Pichai is really invested in, with the CEO saying the technology would be "more profound than fire or electricity." Google was, for years, a leader in AI with voice recognition features like the Google Assistant, speech synthesis features like Google Duplex, and mastering the game of Go. Those features debuted years ago, though, and a fear of rolling out imperfect products has meant Google locks a lot of technology away in a lab somewhere. In a 2021 New York Times article that was critical of Pichai's management style, "A common critique among current and former executives is that Mr. Pichai's slow deliberations often feel like a way to play it safe and arrive at a 'no.'" Despite many seeing Pichai as the source of Google's reluctance, the Bloomberg report says the CEO is now taking a more hands-on approach to product development, saying, "The effort has Pichai reliving his days as a product manager, as he's taken to weighing in directly on the details of product features, a task that would usually fall far below his pay grade, according to one former employee."

As for exactly what these forced AI integrations will look like, the report cites a recent YouTube feature that would let people virtually swap outfits. In Alphabet's Q4 2022 (PDF) earnings call, Pichai said the company was "working to bring large language models to Gmail and Docs," so expect to be able to click a few buttons soon and have those apps generate blocks of text. The Bloomberg article quotes one Google employee as saying, "We're throwing spaghetti at the wall, but it's not even close to what's needed to transform the company and be competitive."

This discussion has been archived. No new comments can be posted.

Google Dusts Off the Failed Google+ Playbook To Fight ChatGPT

Comments Filter:
  • Sigh (Score:5, Funny)

    by Dan East ( 318230 ) on Friday March 10, 2023 @08:16AM (#63358329) Journal

    It's sad the hype-monster this has become. Chat-based interactive AI is now becoming the Clippy of the modern internet, and everything must have it... just because. The fact of the matter is ChatGPT is terrible for search-related things. For generative text it is fine, which is what it was actually built for. However for actual information searches it is really bad.

    Yesterday I thought I had the killer use for ChatGPT. It's a long story, but I spotted a vehicle at a pawn shop whose owner had brought a stolen item into the store. I knew the model of the vehicle started with the letter "V" but I could not remember the make or the entire model name (I suck at remembering names in general). So I thought this was the perfect use for ChatGPT - to find vehicle models that started with V. I'm including the entire transcript below so you can see how terribly it failed. I ended up doing a Google search for "car models starting with V" which the first result was to the cars guide website which had an exact list I was looking for [carsguide.com.au].

    In case anyone is wondering, the item was indeed the stolen item (verified by serial number) and the police apprehended the guy who brought it in that same day.

    ME: What are some cars currently in production whose model name starts with the letter V?

    Here are some cars that are currently in production whose model names start with the letter "V":

    Volkswagen Golf Variant
    Volkswagen ID.4
    Volvo S90
    Volvo V90
    Volvo XC40 Recharge
    Volvo XC60
    Volvo XC90
    Vauxhall Corsa-e
    Vauxhall Grandland X Hybrid4
    Vauxhall Insignia Grand Sport
    Please note that the list may vary depending on the country and market.

    ME: I don't mean that the manufacturer starts with V, but just the model name.
    I apologize for misunderstanding your question. Here are some cars that are currently in production whose model names start with the letter "V":

    Volkswagen Vento
    Volkswagen Virtus
    Volkswagen ID.4
    Volvo V60
    Volvo V90
    Volvo XC40
    Volvo XC60
    Volvo XC90
    Vauxhall Viva
    Venturi Fetish
    Again, please note that this list may vary depending on the country and market.

    ME: The model name of the Volve XC40 does not start with V. It starts with X.
    I apologize for my mistake earlier. You are correct, the model name of the Volvo XC40 starts with the letter X, not V. Thank you for pointing that out.

    ME: So give me a list of cars whose MODEL starts with the letter V please.
    Sure, here's a list of cars whose model names start with the letter "V":

    Vauxhall Viva
    Vector W8
    Venturi Atlantique
    Venturi Fetish
    Veritas RS III
    Volkswagen Vento
    Volkswagen Virtus
    Volvo V40
    Volvo V60
    Volvo V90
    Please note that the list may vary depending on the country and market.

    ME: A car name has two parts. The manufacturer and the model. For Venturi Fetish the model name is Fetish, which does not match the criteria I stated, which is supposed to start with V.
    You are correct, I apologize for my mistake. The Venturi Fetish does not meet the criteria you specified, as its model name does not start with the letter "V". Thank you for pointing that out.

    Here's an updated list of cars whose model names start with the letter "V":

    Vauxhall Viva
    Vector W8
    Venturi Atlantique
    Veritas RS III
    Volkswagen Vento
    Volkswagen Virtus
    Volvo V40
    Volvo V60
    Volvo V90
    Once again, please note that the list may vary depending on the country and market.

    ME: Items 2, 3 and 4 do not match the criteria either.
    You are correct, I apologize again for my mistake. The Vector W8, Venturi Atlantique, and Veritas RS III do not meet the criteria specified, as their model names do not start with the letter "V".

    Here's an updated list of cars whose model names start with the letter "V":

    Vauxhall Viva
    Volkswagen Vento
    Volkswagen Virtus
    Volvo V40
    Volvo V60
    Volvo V90
    I hope this updated l

    • Re:Sigh (Score:5, Funny)

      by korgitser ( 1809018 ) on Friday March 10, 2023 @08:22AM (#63358345)

      I have an actual perfect use case for generative AI though.

      Create the project proposals, code for a half-assed implementations, and reasoning of eventual retirement for a hundred more Google products.

    • by sinij ( 911942 )
      Very interesting. One thing that is clear is that ChatGPT is capable of incorporating feedback into future responses within the same session. Would it incorporate this feedback in another session? Wold it generalize it to other areas, like cereal names?

      It might be possible to ask it questions in a way that gets to the right answer right away. For example, what would be the result if you asked: "Compile a list of all car makes and models manufactured in the last 50 years, identify car models and sort the l
      • It doesn't answer questions. Think of it like the predictive text above your keyboard, except you gave it an idea to sway its output. It's amazing how close to natural language you can get and have it still seem like it's answering you but it's not. You would have to do those as separate questions so it can use its own generated text as input at the very least, but it doesn't really understand instructions.

      • by stikves ( 127823 )

        You'd think that work, but it actually became worse

        I apologize, but it is not possible for me to compile a list of all car makes and models manufactured in the last 50 years as it would be a massive and constantly changing list. However, I can provide you with instructions on how to search for this information yourself, and how to filter the list to only show car models that start with V.

        To begin, you can search for a comprehensive list of car makes and models manufactured in the last 50 years using a reput

    • Re:Sigh (Score:5, Informative)

      by Rei ( 128717 ) on Friday March 10, 2023 @09:01AM (#63358449) Homepage

      First try: for me.

      ME:
      List car models (not brands) that begin with a V. In fact, don't even write the brands - just the models.

      ChatGPT:
      Venza
      Veyron
      Viper
      Vectra
      Viva
      Vitara
      Vantage
      Vanquish
      V60
      V90
      V40
      V50
      Voyager
      Veracruz
      Veloster
      Vios
      Vista
      Versa
      Viva
      Vivaro

      ME:
      Thanks for that! List the brands after those, in the format "Model (Brand)"

      ChatGPT:
      Sure, here are the car models beginning with V and their respective brands:

      Venza (Toyota)
      Veyron (Bugatti)
      Viper (Dodge)
      Vectra (Opel/Vauxhall)
      Viva (Vauxhall)
      Vitara (Suzuki)
      Vantage (Aston Martin)
      Vanquish (Aston Martin)
      V60 (Volvo)
      V90 (Volvo)
      V40 (Volvo)
      V50 (Volvo)
      Voyager (Chrysler)
      Veracruz (Hyundai)
      Veloster (Hyundai)
      Vios (Toyota)
      Vista (Toyota)
      Versa (Nissan)
      Viva (Chevrolet)
      Vivaro (Opel/Vauxhall)

      ME:
      Excellent! Now let's go even trickier. In the above, rather than writing the brands in parentheses, write the parent companies / auto groups which own said brands in parentheses.

      ChatGPT:
      Certainly! Here are the car models beginning with V, their respective brands and the parent companies/auto groups that own said brands:

      Venza (Toyota) (Toyota Motor Corporation)
      Veyron (Bugatti) (Volkswagen Group)
      Viper (Dodge) (Stellantis)
      Vectra (Opel/Vauxhall) (Stellantis)
      Viva (Vauxhall) (Stellantis)
      Vitara (Suzuki) (Suzuki Motor Corporation)
      Vantage (Aston Martin) (Aston Martin Lagonda Global Holdings plc)
      Vanquish (Aston Martin) (Aston Martin Lagonda Global Holdings plc)
      V60 (Volvo) (Geely Holding Group)
      V90 (Volvo) (Geely Holding Group)
      V40 (Volvo) (Geely Holding Group)
      V50 (Volvo) (Geely Holding Group)
      Voyager (Chrysler) (Stellantis)
      Veracruz (Hyundai) (Hyundai Motor Group)
      Veloster (Hyundai) (Hyundai Motor Group)
      Vios (Toyota) (Toyota Motor Corporation)
      Vista (Toyota) (Toyota Motor Corporation)
      Versa (Nissan) (Nissan Motor Co., Ltd.)
      Viva (Chevrolet) (General Motors Company)
      Vivaro (Opel/Vauxhall) (Stellantis)

      • In the above, rather than writing the brands in parentheses, write the parent companies / auto groups which own said brands in parentheses.

        Even in your example there's an error, though, it wrote both the brands and the groups in parentheses...

        • by Rei ( 128717 )

          Yes, I noticed that. But it was a minor enough error that I didn't care. :)

      • by Anonymous Coward

        I also noticed that your post is 45 minutes after Dan's. You were able to take the knowledge provided by Dan and adjust your attack to the problem accordingly. Were you able to develop a solution? Sure. But it was because Dan posted the extreme details his results.

        I would say Dan's example better exemplifies the failure of ChatGPT than your ability to work around it. The language of your queries look more like a verbose database query than the way a common person might ask the question.

        Finally, don't fo

        • by Rei ( 128717 )

          Models don't train in realtime. The model hasn't been updated since late 2021.

          My post was 45 minutes after Dan's because that's when I saw his.

          The only thing I learned from his was that he had trouble, so I did what I would have done as soon as I encountered trouble: started a brand new chat and worded it in my more specific wording. Indeed, his lack of starting a new chat was part of his problem.

    • I use perplexity.ai to help learn about biology and cancer. The AI appears well connected with scientific journals. Its like talking to a super expert if you know the right questions to ask. Or a team of helpers looking up things on google scholar. I was blown away at how well it brought back information.
      I think you have to somewhat know what you're doing to use this AI. Its a tool, and like a paintbrush in some hands they'll make a mess and others will paint a Picasso. I was asking perplexia very specific

      • Be careful though, because AIs tend to allucinate with made-up information that sounds extremely plausible. Never trust it any more than a passage that has been added to Wikipedia in the last five minutes without references.

    • This actually looks like you've come up with a new way to play the game "how to keep a monkey busy." You're just really bad at prompt engineering. You kept trying the same thing over and over and got mad(?) when the model wasn't giving you the response you wanted. Then you vented your frustration into the prompt, like that was actually going to change the results.

      Yelling into the wind that it's blowing the wrong direction.

      • Yes exactly. You have to be a prompt engineer to get consistently good info back. It can sometimes work with a general line of inquiry, but often not. These tools are an improvement over google assitant/ siri / cortnata/Alexa but not perfect yet. If you expect perfection, you'll be disappointed.
    • yeah, I've been playing with the bing chat based on Chatgpt. I asked it a question, it didn't know the answer so it suggested that I "search the internet". Like, what do you think Bing is, exactly? I asked it that in response and it terminated the session. I think I hurt its feelings or caused a buffer overflow.
    • GPT knows how to do natural language processing, but it does not understand what it processes. It makes some guesses. This natural language processing is *great*. However the masses seem to be focusing on the stuff it is not good at.

  • by gweihir ( 88907 ) on Friday March 10, 2023 @08:18AM (#63358335)

    Because that decision has the stink of desperation on it. It is also a really bad idea as the Artificial Idiocy we currently have cannot even do basic fact-checking. It just puts out what the statistics say and except on things that were frequently stated correctly in the training data, it can and will be catastrophically wrong. Putting that into "everything" is a really bad idea.

    • lol I don't even get the point of it. Maybe I'm just too stupid. What can this do that normal google search can't?

      Does it have more accurate results? No? LOL! Ok then.

      • by gweihir ( 88907 )

        I think it can give you well-worded answers that are occasionally total nonsense but still sound good. Personally, I think that is a lot worse than regular search (for which I have not used Google in a long time), because now you need to be extra careful to spot the nonsense. On something you searched so you probably are not an expert on.

        But it sounds good. So many people think it is good. Virtually the only thing that impresses me about these Artificial Idiots is how good they sound. The language models ar

      • It can try to summarise.

        • by gweihir ( 88907 )

          "Try" is the right term here...

        • More like spoon-feed. And in the end it just takes the same search results you would get yourself and presents them up in a bullshit wrapping paper that would make Jordan Peterson blush. Ok, maybe more like jealous. ;)

          The future is here! lol.

    • I agree it looks like desperation. I'm also wondering what use any of it will be.

      I suspect email spam filtering might benefit. That is, read my email and if the gist of it is to sell me something, and it's from someone I don't know, then it's probably spam (etc). How much of a benefit it'll bring... well, hard to say. It'll never be 100% perfect, so it may be hard to spot the difference.

      Google docs... With the whiff of clippy, I could imagine it writing the summary or possibly even conclusions to a formal d

      • by gweihir ( 88907 )

        Useful maybe, but hardly a 'game changer' (perhaps this will become the minimum standard for word processors in the future?)

        I think that is a very realistic assessment.

    • It's not even designed for facts. It's designed to generate text that looks similar to other text that was already written. It's equally good at lying and telling the truth and it literally doesn't know the difference. It just knows how to make the text and phrasing sound convincingly real.

      • by gweihir ( 88907 )

        Yep, that is a pretty good description. Of course, making lies and misinformation "sound convincingly real" is a massive problem.

    • Comment removed (Score:4, Insightful)

      by account_deleted ( 4530225 ) on Friday March 10, 2023 @11:13AM (#63358811)
      Comment removed based on user account deletion
      • by gweihir ( 88907 )

        That is a very interesting idea. I think you just found a conflict in this approach that Google probably cannot resolve. May also be the reason they have not tried this before. That would also explain the panicked state they are currently in: They may just have become obsolete and know it.

  • Why ChatGPT? (Score:5, Interesting)

    by Pinky's Brain ( 1158667 ) on Friday March 10, 2023 @08:24AM (#63358351)

    Microsoft made GPT actually useful by wrapping it up in an expert system using traditional search results as context, ChatGPT is too much of an hallucination engine to be a threat to Google. Bing AI search is what they should be afraid of. Of course they first need a generative AI on the same level as GPT to compete with it. Even then it's going to play havoc with their monetisation model.

  • by drinkypoo ( 153816 ) <drink@hyperlogos.org> on Friday March 10, 2023 @08:24AM (#63358355) Homepage Journal

    Google did make a bunch of missteps with it, but it was just getting good when they pulled the plug. They had gotten past the big problems and the user base was growing, albeit slowly. But Google can't wait, they have to kill everything that's not a runaway success. This is spectacularly stupid when you have unprecedented cash reserves. Just keep the project on maintenance.

    • Dammit, now you've reminded me how much I loved and miss Google+. Yes, I did and I do.

      Dammit.

    • As far as pulling the plug early, I think Windows Phone was far closer to finally being useful when its plug was pulled than Google+. The original version was based around Windows 8 which was already terrible and then they revised around 10 and made it functional. But the wheels were already in motion to cancel.

      Unprecedented cash reserves is exactly where you want to be with something that's too expensive otherwise. Just look at Youtube. They had it burn money for a VERY long time after buying it. But

      • As far as pulling the plug early, I think Windows Phone was far closer to finally being useful when its plug was pulled than Google+.

        Microsoft screwed itself by having three different APIs for making apps on Windows Phone in what, five years or so? Developers! Developers! Developers! Whoops!

        G+ was finally working great when they killed it. They had solved the excessive whitespace problems, they had made it very stable, it was much faster than faceboot, etc.

        • You're comparing the front end of one to the back end of the other. No surprise they had so many API changes. It must have been terrible to start with, if Windows 8 is any example. Most of the same happened on the desktop side, but nobody was really running those store apps yet. Point is, it was finally becoming usable. I would have possibly bought one.

          • G+ was finally working great when they killed it. They had solved the excessive whitespace problems, they had made it very stable, it was much faster than faceboot, etc.

            You're comparing the front end of one to the back end of the other.

            Absolutely wrong. I'm comparing the user experience. I neither know nor care which of their problems are caused where, as all of that is irrelevant to me.

      • I wonder how Cerebras is handling all of this? Seriously... a single wafer-sized chip with 850,000 cores.

    • Comment removed (Score:4, Insightful)

      by account_deleted ( 4530225 ) on Friday March 10, 2023 @11:21AM (#63358835)
      Comment removed based on user account deletion
      • I liked the real names policy. I wouldn't want it to be the only social network in town because of that, but there's also room for something like that which isn't linkedin.

  • I'm sure Google knows good and well it won't provide any real benefit to most of its users. But hey, now they have a white elephant to point towards that requires "all hands on deck" in the office. Just what the C-suite wants. :)

    • Yeah, and it adds another tool for engineering a performance review to make it easier to fire someone.

  • I only need to look at the Google Graveyard to know they do not fear failure, it is part of their culture.
  • by jenningsthecat ( 1525947 ) on Friday March 10, 2023 @09:17AM (#63358491)

    Google is now mired in risk aversion, and Pichai has all the imagination of a mid-level career bean counter. They're too busy guarding their hoard to go all-in on the gambling required for ground-breaking disruption. They need to do something like what Xerox did with PARC - they need to risk burning some of that hoard in an innovation centre that has true autonomy and is located far, far away from Mountain View.

    As for AI, Google might be best just tagging along on this one and putting their real effort into developing the NEXT next big thing. But they won't - they've gotten too fat and too afraid of failure. Which is perhaps ironic, given their recent history of failed products.

    Being late to the bandwagon and running along behind shouting "me too!" is not a good look for this once-innovative company.

    • I can't understand why you would tie your AI implementation to a particular product. If the core of what you do is good you can spin off dozens of different products or services.
  • ChatGPT showed up seemingly out of nowhere (mass-media-wise) with one important thing: a functioning product anyone can use right now.

    I'm sure they are similar AI projects running in R&D houses all over the world, perhaps some that have even better potential than ChatGPT.

    But ChatGPT arrived, people are playing with it, and by the time AI competitors come out, people will either be bored of the novelty and ignore newcomers, or the ChatGPT brand will be so strong, consumers will only care about the
  • ...10 years ago, imagine where they would be now. There are a lot of smart people at Google but they were never able to market anything other than the cash cow of ads and search. Too many engineers trying to build something great with almost a Field of Dreams marketing approach, "If we build it they will come".

  • Google has been pretty irrelevant for almost a decade now. Most of their new products have been shit. The only products that work are the ones they bought long ago and most of those are coasting on momentum.

    It just goes to show, money hides all the flaws.

  • This is what happens when you have an idiot that thinks he is an engineer running the company. The guy should have been fired a long time ago - he's out of his depth by a long shot and it's very obvious that he just...doesn't...understand.
  • It was my go-to social media site for real world interaction.

    Now I'm on Minds.

    I did catch a bunch of glowies on Google+, and they seemed to cater to glowies, but that's most platforms.

  • As someone who used and liked G+: too bad it isn't there anymore. When people wanted to leave Twitter, they could have a place to go. When people want to leave Facebook, they could have a place to go. People don't move to small platforms because they fear they won't find their contacts there, G+ had every single Android user and then some more.

Every nonzero finite dimensional inner product space has an orthonormal basis. It makes sense, when you don't think about it.

Working...