Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Google

Google's Best Gemini Demo Was Faked (techcrunch.com) 49

Speaking of early-impressions of Gemini, users' confidence in Google might be shaken further to learn that the company pretty much faked the most impressive demo of Gemini. TechCrunch: A video called "Hands-on with Gemini: Interacting with multimodal AI" hit a million views over the last day, and it's not hard to see why. The impressive demo "highlights some of our favorite interactions with Gemini," showing how the multimodal model (that is, it understands and mixes language and visual understanding) can be flexible and responsive to a variety of inputs.

To begin with, it narrates an evolving sketch of a duck from a squiggle to a completed drawing, which it says is an unrealistic color, then evinces surprise ("What the quack!") when seeing a toy blue duck. [...] Just one problem: the video isn't real. "We created the demo by capturing footage in order to test Gemini's capabilities on a wide range of challenges. Then we prompted Gemini using still image frames from the footage, and prompting via text." So although it might kind of do the things Google shows in the video, it didn't, and maybe couldn't, do them live and in the way they implied. In actuality, it was a series of carefully tuned text prompts with still images, clearly selected and shortened to misrepresent what the interaction is actually like.

This discussion has been archived.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

Google's Best Gemini Demo Was Faked

Comments Filter:
  • Do no evil, assholes.

    You forgot where you (claim) to have come from. You survive only on inertia now.

  • by willkane ( 6824186 ) on Friday December 08, 2023 @10:05AM (#64066361)
    Looks like Google for the first time in more than 20 years is feeling a little of pressure.

    (* Mechanical Turk [wikipedia.org])
  • Yo, AI. (Score:3, Insightful)

    by Anonymous Coward on Friday December 08, 2023 @10:07AM (#64066367)

    Do we hear a specific accent in the video? When we hear this accent, would you expect that we get a presentation full of exaggerations and bullshit?

  • by davidwr ( 791652 ) on Friday December 08, 2023 @10:18AM (#64066381) Homepage Journal

    It's a faaaaake [youtu.be].

  • Whooops (Score:4, Interesting)

    by timeOday ( 582209 ) on Friday December 08, 2023 @10:25AM (#64066399)
    I can see how making up some dialogue for some footage is a normal thing to do in most contexts, but in this case how the tool responds to the interaction is the whole point. What's the endgame here? People are going to interact with this in person so if they're overclaiming, people are going to know.

    ChatGPT was the opposite of this. People didn't first see it through ads. They pulled up the URL, and were amazed.

    • Because people who make decisions are too important to waste time with the these sort of details, that’s for peons, scapegoats and executives that don’t meet quarterly projections.
    • ChatGPT was the opposite of this. People didn't first see it through ads. They pulled up the URL, and were amazed.

      Well there was the sparks of AGI paper and video presentation... several things in that paper were not reproducible including the svg unicorn drawing once GPT-4 was publicly released. Probably due to RLHF induced brain damage applied to avoid bad press.

  • Honestly reporting the progress made would have been very impressive, but someone evidently believed it needed more
    The evolution of AI should be a series of technological advancements, honestly reported by the tech press
    Instead, it has turned into a cutthroat competition for investment dollars and the attention of the general public, by any means necessary

    • A manager probably demanded more and the company culture is probably diseased where avoiding confrontation with management is the common way to respond. Better to lie to the customer than say no to the manager ...

      It's a cutthroat competition to suck up to the boss and ride the sinking ship down.

    • by jvkjvk ( 102057 )

      >The evolution of AI should be a series of technological advancements, honestly reported by the tech press

      When has this been true of any technological advancement? Please let me know the golden era where this was happening.

  • An analysis of part of the demo has proven that it was created using AI.
  • by ttyler ( 20687 ) on Friday December 08, 2023 @10:50AM (#64066455)
  • It's not necessarily that stuff doesn't work... it's a demo

    Most companies want to avoid scenarios where the computer crashes mid-demo like Windows crashing on Bill Gates

    Now how good it is... time will tell

    • by Mascot ( 120795 )

      If the summary is correct, then this wasn't just a case of "it crashes a lot so we made a video that's spliced together from multiple takes". That would be something like what you're talking about, in my view. That would be an actual demonstration of a capability that exists, even if not stable enough to dare show live.

      This was more like "let's show a video of us feeding live video and audio to an AI, with the AI narrating what it sees and responding to verbal conversation about what's going on in the video

      • by Mascot ( 120795 )

        To add to my above comment, I decided to click through to the article and check the video itself. To Google's credit, they do explain a few things in a link in the video description. They also apparently responded to criticism with "The video illustrates what the multimodal user experiences built with Gemini could look like. We made it to inspire developers."

        If they had named the video something like "here's what we hope Gemini will eventually do", that would have been perfectly fine. But they didn't. They

      • So, they Googled it.
    • by jvkjvk ( 102057 )

      >It's not necessarily that stuff doesn't work... it's a demo

      There's a difference between a demo and a scripted presentation. A BIG difference. Many companies have been pulled up and chastised for giving scripted presentations where they said they were giving demos.

      >Most companies want to avoid scenarios where the computer crashes mid-demo like Windows crashing on Bill Gates

      Yes, well, that's the whole point, or part of it. Is the thing good enough to actually do what they are claiming, as seen live on

  • I hate to break it to you guys, but Superbowl halftime shows are lip synched, too.

    "Breaking news: Demo videos are edited to maker products look better, film at 11."

    • by jvkjvk ( 102057 )

      This doesn't seem so much like a demo as a scripted presentation. There is a big difference, and Google is touting this as if it was a demo, which again, it does not seem to fit the description of.

    • by Junta ( 36770 )

      There's a difference between editing together a representative video of having the bugs worked out that came up during attempted recording, or taking several prerecorded runs at a demo to getting it right, and outright fabricating a video implying function that doesn't exist even in theory.

      The concept video they imagined misrepresents the capability, by stitching together independent examples that they retroactively decided could fit a neat narrative if represented as a contiguous interaction in real time c

      • The concept video they imagined misrepresents the capability, by stitching together independent examples that they retroactively decided could fit a neat narrative if represented as a contiguous interaction in real time conversation. Likely skipping over more verbose prompts and refinements made after a disappointing response, and outright discarding attempts that landed nowhere.

        This literally describes editing to make the product look better.

        It would be like if I showed up on stage and lip synced to Frank Sinatra and claimed it was my voice to get people to hire me to sing at events.

        This is not the least bit analogous to what you described happening. The more accurate analogy would be Frank Sinatra editing together parts of different takes of his own singing and promoting it as an example of his live show. Which, I hate to break it to you, is how a lot of albums are made.

  • by Guignol ( 159087 ) on Friday December 08, 2023 @12:21PM (#64066701)
    Google has denied engaging in disinformation campaigns...
    • by sinij ( 911942 )
      Checks out. I just Googled and there was nothing found.
    • Sorry about that, chief! I hope I wasn't out of line with that hallucination about the old realtime duck trick! Would you believe that I can beat ChatGPT4? No? Would you believe I can beat ChatGPT3.5? What about ChatGPT2.11? Word for Windows? DOS 3.1? Ok ok please don't tell me to leave! I asked you not to tell me that!
  • by Motleypuss ( 10291831 ) on Friday December 08, 2023 @01:40PM (#64066891)
    It's the Mechanical Turk all over again.
  • Whatever the fuck this I don't care I saw "Google" and my brain said "Fuck whatever comes next". What Google does best these days is create short-lived, often faked, products that fail soon afterwards and get withdrawn and covered up by the company. Well, that is, when they aren't too busy sharing my data with corporate fascists or the brain-dead government. Google I hope you dry up and die because it would be so deserved.

    What I do care about is that they are stomping on another project called Gemini [geminiprotocol.net] wit
  • There is no AI, it's just what this round bound-to-fail .com startups are calling if/else trees too complicated to attempt to debug.

When you are working hard, get up and retch every so often.

Working...