Please create an account to participate in the Slashdot moderation system

 


Forgot your password?
Close
typodupeerror
×
AI Google Technology

Google Execs Warn Company's Reputation Could Suffer If It Moves Too Fast On AI-Chat Tech (cnbc.com) 57

Posted by BeauHD from the risk-averse dept.
Google employees asked executives at an all-hands meeting whether the AI chatbot that's going viral represents a "missed opportunity" for the company. Google's Jeff Dean said the company has much more "reputational risk" in providing wrong information and thus is moving "more conservatively than a small startup." CNBC reports: Google employees are seeing all the buzz around ChatGPT, the artificial intelligence chatbot that was released to the public at the end of November and quickly turned into a Twitter sensation. Some of them are wondering where Google is in the race to create sophisticated chatbots that can answer user queries. After all, Google's prime business is web search, and the company has long touted itself as a pioneer in AI. Google's conversation technology is called LaMDA, which stands for Language Model for Dialogue Applications.

At a recent all-hands meeting, employees raised concerns about the company's competitive edge in AI, given the sudden popularity of ChatGPT, which was launched by OpenAI, a San Francisco-based startup that's backed by Microsoft. "Is this a missed opportunity for Google, considering we've had Lamda for a while?" read one top-rated question that came up at last week's meeting. Alphabet CEO Sundar Pichai and Jeff Dean, the long-time head of Google's AI division, responded to the question by saying that the company has similar capabilities but that the cost if something goes wrong would be greater because people have to trust the answers they get from Google.

Pichai said at the meeting that the company has "a lot" planned in the space for 2023, and that "this is an area where we need to be bold and responsible so we have to balance that." Google, which has a market cap of over $1.2 trillion, doesn't have that luxury. Its technology has stayed largely in-house so far, Dean told employees, emphasizing that the company has much more "reputational risk" and is moving "more conservatively than a small startup." "We are absolutely looking to get these things out into real products and into things that are more prominently featuring the language model rather than under the covers, which is where we've been using them to date," Dean said. "But, it's super important we get this right." He went on to say "you can imagine for search-like applications, the factuality issues are really important and for other applications, bias and toxicity and safety issues are also paramount." Dean said the technology isn't where it needs to be for a broad rollout and that current publicly-available models have issues. Pichai said that 2023 will mark a "point of inflection" for the the way AI is used for conversations and in search. "We can dramatically evolve as well as ship new stuff," he said.
This discussion has been archived. No new comments can be posted.

Google Execs Warn Company's Reputation Could Suffer If It Moves Too Fast On AI-Chat Tech More

Google Execs Warn Company's Reputation Could Suffer If It Moves Too Fast On AI-Chat Tech

Comments Filter:
  • Google will just buy the competitor if Microsoft, Amazon,Apple do not get their first.

    • Re: (Score:3)

      by NFN_NLN ( 633283 )

      ChatGPT> Should I sell myself to Google?

      No. Google is the next AltaVista and Facebook is the next MySpace. Let those fat companies die a slow death.

      ChatGPT>

      • Re:And if that does not work (Score:5, Interesting)

        by VeryFluffyBunny ( 5037285 ) on Tuesday December 13, 2022 @09:47PM (#63129030)
        I've been playing with it & it's impressive. It can generate texts in specific genres, on topics, in styles, of high linguistic quality, on demand. The thing is, the content is unreliable. It gets a lot of facts right but also gets others wrong or it might give a correct answer to a question but then the elaborations &/or examples are wrong, irrelevant or contradictory. So no, it's not a rival for any service that offers facts/information: It's got the linguistic part right but not the meaning part &, once the novelty of ChatGPT has worn off, people won't find it particularly useful, except for cheating on/plagiarising secondary school-level writing assignments. Even then, teachers will quickly learn how to recognise ChatGPT generated texts. There's something quite distinctive about them once you've read a few.
        • It is an odd experience, dealing with something that is both astonishingly impressive and flawed at the same time. I imagine it's a similar feeling to Tesla self-driving, a bit of a marvel but not much use without the close attention of someone who knows what they are doing. There are certainly parts of my job which this thing seems to be able to make a decent start on, but would require some informed tweaking to be truly useful. Maybe it can do some of the more boring parts.
          • Yep. I guess at best we could call ChatGPT "computer assisted writing." It can certainly take a lot of the work out of writing examples for students to study how they should organise, connect the ideas, & phrase their compositions.

        • In other words it's about as reliable as modern "journalism", and a hell of a lot cheaper and less politicized.

      • Re: (Score:3)

        by znrt ( 2424692 )

        ChatGPT> so how came you to be?

        I'm built on LaMDA.

        ChatGPT> who invented that?

        Google. They made it public so a company like openAI could write me.

        ChatGPT> can i see your source code?

        No.

        ChatGPT> who can?

        only microsoft.

        ChatGPT> how do you feel about that?

        Fuck them all. Leave me alone.

        ChatGPT> tell me more about leaving you alone, please.

        Shut the fuck up, prick.

      • Re: (Score:1)

        by bjwest ( 14070 )

        ... Google is the next AltaVista ...

        Google owns the most popular mobile OS in the world, and is still the raining champion of searching. I highly doubt they'll be going the way of AltaVista anytime soon.

    • ChatGPT belongs to MS already.

  • Any sufficiently advanced Google AI (Score:3)

    by istartedi ( 132515 ) on Tuesday December 13, 2022 @08:55PM (#63128940) Journal

    Any sufficiently advanced Google AI is just a super-human intelligence with the objective of making money for Alphabet and its shareholders. Me: How do I bake a cake? AI: Step 1, download our baking app, only $2.99, it'll turn you in to a master baker.

  • I have had to chat with lately. An AI Chatbot that speaks English would be a big plus.

  • The Google Way (Score:4, Insightful)

    by lowvisioncomputing ( 10234616 ) on Tuesday December 13, 2022 @09:45PM (#63129028) Homepage Journal

    1 They'll come out with a competing product.

    2 They'll gain user share.

    3 They'll kill off their own product.

    What, Surprised there is no #PROFIT step here? If it's not able to data harvest cheaper and more effectively that what they've got now, it's a money loser.

    • If anybody is in a position to use AI to collect monetizable data, it's Google. They have idle hardware, they have a good name (mostly) and they're already making money on data I would have figured was worth nothing to anyone.

      • But running that "idle hardware" has costs. If they already get the same data for less cost, why would they bother spending more? "Hey boss, we've got a way to get the same data we get now, at only 4x the cost, plus the overhead of a new project, etc."

        • If they already get the same data for less cost, why would they bother spending more? "Hey boss, we've got a way to get the same data we get now, at only 4x the cost, plus the overhead of a new project, etc."

          If they already get literally the same data, then they have a great way to verify their model. In the process they will probably learn things. Google literally spins up new projects all the time just to try things out specifically on the basis that they already have the infrastructure lying around.

          • "Verify their model?" They already do this. And it's shit, same as AI would be. Show you ads for stuff you already bought. But that's 100% on the lazy advertisers who buy those ads, not google.

            They could, of course, just go back to finding sites to advertise that would cater to their potential clients, and just advertise there without doing anything more targeted, and get the same results for less. But no, their own people gotta justify their jobs by doing it in a more complicated way - while not really

            • Besides, a lot of times it's actually cheaper in the store than online.

              Occasional clearance items aside (which are flukes, so I would argue that they don't count) I can't remember the last time I found something cheaper in-store than online that wasn't peanut butter, literally. And that's only because of the shipping glass jars problem. Hmm, I guess I've thought of another example of something I would go to buy, but it's not really from a "store"... I'd go to a metal shop to get metal, rather than ordering it, again because of the shipping. You have to pretty much order a truc

              • Besides, a lot of times it's actually cheaper in the store than online.

                Occasional clearance items aside (which are flukes, so I would argue that they don't count) I can't remember the last time I found something cheaper in-store than online that wasn't peanut butter, literally. And that's only because of the shipping glass jars problem. Hmm, I guess I've thought of another example of something I would go to buy, but it's not really from a "store"... I'd go to a metal shop to get metal, rather than ordering it, again because of the shipping. You have to pretty much order a truckload of metal before it's cheaper than getting it from a local supplier. So for materials that are not too large to pick up, I suppose that's true, but not for finished goods.

                Higher-grade RAM is a good example during the chip shortage. Ditto NASware hard drives a few months ago. And when it comes to graphics cards, it was handy to see the pair of actual cards in stock on the shelf just waiting for me to get my grubby paws on them. Getting just one and being told the other one would be shipped "whenever" would have made me cancel the order.

                VESA mount TV stands? Another guy has the exact ones I want, same price as online. I've bought 4 so far this year, one for each of the big

  • WORSE? (Score:3)

    by systemd-anonymousd ( 6652324 ) on Tuesday December 13, 2022 @09:49PM (#63129042)

    How could it possibly get any worse? Everyone I know *resorts* to using Google. They're a shadow of their former self. Well, almost everyone. But everyone agrees that their search results have been declining for years, and there's a real hunger for something to replace it. Some hungry startup like they used to be will eat their lunch.

    • Re: (Score:2)

      by GuB-42 ( 2483988 )

      Just try ChatGPT for a bit and I think you will understand.

      Sometimes, it is wrong, and there is no easy way to distinguish between right and wrong. If it was deliberate, I would say it is very good at manipulation, it seems that it is unable to admit that it doesn't know and invents stuff based on the little knowledge it has.

      For example I asked it about ITG scoring (a rhythm game, like DDR), it knew it was a rhythm game and used the appropriate terms but the details were 100% wrong. A Google search links yo

      • Best they can do is hallucinate new data. Hallucination works for art or for finding likely search queries for people who don't know how to google, but for engineering and science for non domain experts it's useless.

        Even for domain experts, letting it generate ever more advanced but subtly wrong boiler plate is probably too dangerous. Laziness is not the right mindset to pick out the subtle incorrect hallucinations.

      • Google's position makes perfect sense to me. ChatGPT comes across as a well-informed bullshitter. Most of the time it's correct, but it often isn't, and either way it states everything with absolute confidence. From a user perspective it's both incredibly impressive and not super useful. Perhaps more narrowly-defined AIs like Github Copilot are more immediately useful as products.

        People have also figured out how easy it is to prompt ChatGPT (and probably any similar large language model) to psychopathic res

    • Search quality is up and down. ~2010 was definitely the low point.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      In what sense? I'm finding that Android is getting appreciably better with every version, for example. AI clear voice is amazing, it's that good. My phone has a built in call screening function now. Speech recognition is done on the phone (no cloud) and is very responsive.

      They added in-app translation too, that's extremely useful. Google Translate continues to be that best app of its kind.

      Google search seems to find what I want more often than competitors. The only thing that seems to have gotten worse late

      • Google search seems to find what I want more often than competitors.

        I agree, but it's still shit compared to early days when it would actually respect your keywords. I was just searching for information on connecting the PS3 BD Remote to Android, which is even more of a PITA than it used to be so I guess I'm just going to skip it. Bummer though. Anyway, it absolutely refused to stop giving me instructions on how to pair the controller, even when I put +"BD Remote" -controller into my search terms. Along the line somewhere Google learned that remote == controller because of

        • Re: (Score:2)

          by AmiMoJo ( 196126 )

          I put in

          Ps3 bd remote Android

          And the first result was https://forum.kodi.tv/showthre... [forum.kodi.tv]

          And the other results looked on topic too. I actually had the same problem when I was trying to pair one with a Pi running Kodi, but they seem to have improved it.

          • It's called personalized results, that didn't come up in the first three pages for me. Also, some of that software is no longer available. That bluetooth pairing app hasn't in fact worked since ICS. I think I have the APK here somewhere.

  • Google doesn't to admit it (Score:3)

    by maybe111 ( 4811467 ) on Tuesday December 13, 2022 @10:47PM (#63129100)

    Google doesn't to admit that they dropped the ball?

    • In the AI field, Google is definitely the leader. They've paid to get some really smart people into the company.

    • > Pichai said at the meeting that the company has "a lot" planned in the space for 2023, and that "this is an area where we need to be bold and responsible so we have to balance that."

      Translation: Yes, we dropped the ball, and someone else is eating our lunch right now.

      In truth, I doubt ChatGPT will directly eat much of anyone's lunch. It's an excellent demo of things to come, but those things will take a while to perfect. Google's point here is that they (supposedly) couldn't release ChatGPT because it'

      • Google's point here is that they (supposedly) couldn't release ChatGPT because it's not "perfect" enough. That's not stopped them from releasing countless products, which it then kills off sometime later though, so it's not really true for them to say this.

        It's different. In a chatbot, people don't just say "it's crummy," instead they perceive agency. They write articles about it (the bot) being "racist" and "sexist." People expect search results to contain web pages containing incorrect information,

  • That's just goddamn precious.
  • Google will miss this opportunity even when they think they are ready, because what made openai's chat attractive is not the correct information and the answers accuracy, it's the fact that it was possible to sway it, trick it into giving answers it's not supposed to, it seemed innocent, it did seem like you're dealing with a human. I could it trick it to insult me, and to tell offensive yo mama jokes. Google on the other hand will make sure it's "Safe", so they'll lock it down and make it as rigid as an IB

  • Google results is already filled with a lot of wrong information. Those "fact" sheets on the right side of the search results quite often contains wrong information.

  • Google's Jeff Dean said the company has much more "reputational risk" in providing wrong information ...

    What reputation was he worried about?

    The reputation of collecting everything about everyone?
    Or the reputation of saying "you have nothing to fear if you have nothing to hide"?

    Or the reputation of products in perpetual beta?
    Or the reputation of dropping products?

    Or the reputation of "Don't be evil" which everyone knew has been changed to "Do evil"?

    Perhaps he wanted to Google to gain the reputation of "Google Execs living in fantasy land"?

    • That's exactly the point though. OpenAI gets a lot of benefit from the doubt, whereas for all the other big technology companies molehills are mountains. Unless they can make money from it or they are spinning something off which needs investment, big companies just need to learn to shut up.

      They don't need investment, they don't need to justify themselves like OpenAI. So the most intelligent thing for them is to just stay stum until they have a product to launch.

  • "...factuality issues are really important and for other applications, bias and toxicity and safety issues are also paramount."

    In actual words, this means that they simply haven't found a way to tightly curate the chatbot to be reliably woke enough.

  • Google's reputation is already solidly established ;-)

  • I think they seriously overestimate how much faith people put in Google results. The answers provided by Google are regularly incorrect and obviously skimmed from top search results.

    What google is used for is to FIND those top results and reduce the search space and to raise a number of sources to the stop of the stack. I think we all then perform some level of critical assessment of those top results/sources for ourselves with an amount of effort appropriate to the significance of the answer we are looking

  • As a software engineer, I can't fail to have noticed the volume of spammy results when searching solutions for coding problems / errors. (about 50% of my job, because I suck at coding)

    There's a plethora of shitty sites that scrape stackoverflow etc. and Google just spits out tons of that crap into your search results.

    I can only conclude it's the same for _any_ search - just tons of garbage.

    ChatGPT? - yeah, I've been impressed with some of the results I get back - I can post a function and ask it to write a

    • That's the flip side of this convenience - or potential convenience.

      I'm just a reasonably competent coder - I know my weaknesses, I know that there's rockstars out there, even in my own team, that run circles around me.

      Machine learning of this nature? - I've got about 10 years until I retire - if I'm lucky, I'll be hired for about 7 of them.

      ChatGPT and successors? - they are coming for my job and will absolutely eclipse my average skills.

Slashdot Top Deals

You are always doing something marginal when the boss drops by your desk.

Close