Google's 'Bard' chatbot completely botches first demo

AppleInsiderAppleInsider
Posted:
in General Discussion edited February 2023
Google recently demonstrated a new ChatGPT rival called Bard, but it made a big factual error in one demo that the company shared on Twitter.

Google shows off its
Google shows off its "Bard" search assistant


Google confirmed on Monday that it was working on an "experimental conversational AI service" called Bard, to compete with the ChatGPT chatbot. The company showed off its capabilities on Tuesday at an event in Paris.

Bard is a chatbot powered by AI that is intended to be used in searches, similar to Microsoft's recent announcement of integrating ChatGPT into Bing.

{"@context":"https://schema.org/","@type":"VideoObject","name":"Google presents : Live from Paris","description":"We're reimagining how people search for, explore and interact with information, making it more natural and intuitive than ever before to find what you need. Join us to learn how we're opening up greater access to information for people everywhere, through Search, Maps and beyond.","thumbnailUrl":"https://i.ytimg.com/vi/yLWXJ22LUEc/sddefault.jpg","uploadDate":"2023-02-08T15:26:02Z","duration":"PT38M18S","embedUrl":""}

Google intends for Bard to enhance its search product similarly, handling questions and answers using natural language. It will condense complex data so users can quickly get the answers they need and learn more.

At one point, though, Bard made an error. In a GIF that Google posted to Twitter, an example showed a person asking, "What new discoveries from the James Webb Space Telescope (JWST) can I tell my 9 year old about?"

Bard presented three bullet points of information, including one that states that the telescope "took the very first pictures of a planet outside of our own solar system."

But as astronomers pointed out, the JWST didn't take the first image of an exoplanet, as NASA shares on its website.

"This highlights the importance of a rigorous testing process, something that we're kicking off this week with our Trusted Tester program," a Google spokesperson told Reuters. "We'll combine external feedback with our own internal testing to make sure Bard's responses meet a high bar for quality, safety and groundedness in real-world information."

Google announced updates to some of its other products, too. For example, Google Lens will let people use an Android device to search what they see in photos and videos across websites and apps they currently use.

Google Translate will also give users additional contextual translation choices with explanations and several examples in the target language. Finally, Google Maps will offer glanceable directions and views of places that users want to visit using augmented reality.

Google Bard availability

Currently, Bard is in a closed beta for testing, and greater public availability will arrive in the coming weeks.

Meanwhile, Microsoft's ChatGPT integration into Bing desktop search is available in a limited preview. Users can try a limited number of queries and sign up for full access in the future.

Read on AppleInsider

Comments

  • Reply 1 of 20
    waveparticlewaveparticle Posts: 1,497member
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
  • Reply 2 of 20
    xiao-zhixiao-zhi Posts: 113member
    Google’s ChatBot is probably accurate enough for Android users.
    lolliverbaconstangravnorodomwatto_cobrajony0
  • Reply 3 of 20
    XedXed Posts: 2,921member
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    The Hubble telescope did it 23 years ago with HD 209458 b. It's 159 light years away. It was also able to directly detect the exoplanet's atmosphere and survey its makeup. 
    edited February 2023 mike1roundaboutnowdewmesphericlollivertmaybaconstangFileMakerFellerravnorodomwatto_cobra
  • Reply 4 of 20
    DarkMouzeDarkMouze Posts: 38member
    waveparticle said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    As noted in the NASA link in the article itself, the first planet directly imaged by a telescope was in 2004, when the VLT telescope pictured one in a system 230 light years away. The mistake referenced in the article was not that they could see a planet using the telescope, but that they were nowhere near being the first to do so.
    lolliverStrangeDaysuraharaFileMakerFellerravnorodomwatto_cobrabyronljony0
  • Reply 5 of 20
    XedXed Posts: 2,921member
    DarkMouze said:
    waveparticle said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    As noted in the NASA link in the article itself, the first planet directly imaged by a telescope was in 2004, when the VLT telescope pictured one in a system 230 light years away. The mistake referenced in the article was not that they could see a planet using the telescope, but that they were nowhere near being the first to do so.
    I believe VLT (Chile) is the first land-based telescope to take an image of an exoplanet.
    lolliverDarkMouzeFileMakerFellerwatto_cobrajony0
  • Reply 6 of 20
    sbdudesbdude Posts: 293member
    waveparticle said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    Did you click the embedded NASA link in the article, or did you substitute your own facts instead?
    lollivermuthuk_vanalingamStrangeDaysbaconstanguraharaFileMakerFellerwatto_cobra
  • Reply 7 of 20
    ihatescreennamesihatescreennames Posts: 1,980member
    AppleInsider said:

    "This highlights the importance of a rigorous testing process, something that we're kicking off this week with our Trusted Tester program," a Google spokesperson told Reuters. "We'll combine external feedback with our own internal testing to make sure Bard's responses meet a high bar for quality, safety and groundedness in real-world information."

    Read on AppleInsider
    Am I reading this right? It seems the Google spokesperson is saying that the job of. Trusted Testers is to verify the information they receive is accurate? I mean, I get that, but is that a paid position or are they expected to do it for free?
    watto_cobra
  • Reply 8 of 20
    bulk001bulk001 Posts: 797member
    From the headline I thought it got 1+1 wrong. Seems that a product that is in testing is bound to get it wrong. You don’t run headlines like”Appleinsider completely butchered the English language” when a typo slips in. 
    gatorguywatto_cobraavon b7byronljony0
  • Reply 9 of 20
    timmilleatimmillea Posts: 261member
    This is the problem with 'big data' type AI, it is as fallible as the underlying data. The process of curating the data is, so far, infeasible. Junk in, junk out. 
    danoxwaveparticleFileMakerFellerwatto_cobra
  • Reply 10 of 20
    danoxdanox Posts: 3,478member
    That exoplanet picture is nice, but most people are looking for something with a hell of a lot more resolution. I can’t wait until we can get more telescopes in space and make a telescope array out of them. That will be really interesting.
    watto_cobrabyronl
  • Reply 11 of 20
    uraharaurahara Posts: 733member
    danox said:
    That exoplanet picture is nice, but most people are looking for something with a hell of a lot more resolution. I can’t wait until we can get more telescopes in space and make a telescope array out of them. That will be really interesting.
    Is this something what being developed? 
    How are they going to deal that the distance between telescopes will be constantly changing because of the curvature of the earth. Or the error is neglectable? 
    watto_cobra
  • Reply 12 of 20
    uraharaurahara Posts: 733member
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    Are you one of these guys?

    A rocket will never be able to leave the Earth’s atmosphere.” — New York Times, 1936

    FileMakerFellerravnorodomwatto_cobrabyronl
  • Reply 13 of 20
    DarkMouzeDarkMouze Posts: 38member
    Xed said:
    DarkMouze said:
    waveparticle said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    As noted in the NASA link in the article itself, the first planet directly imaged by a telescope was in 2004, when the VLT telescope pictured one in a system 230 light years away. The mistake referenced in the article was not that they could see a planet using the telescope, but that they were nowhere near being the first to do so.
    I believe VLT (Chile) is the first land-based telescope to take an image of an exoplanet.
    That makes sense to me…I’m betting you’re right. But that’s certainly not what the NASA piece says. I can see the AI headline now. “NASA Completely Botches Astronomy!”
    gatorguy
  • Reply 14 of 20
    waveparticlewaveparticle Posts: 1,497member
    urahara said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    Are you one of these guys?

    A rocket will never be able to leave the Earth’s atmosphere.” — New York Times, 1936

    This is false news. Rocket has not been invented in 1936. Nazi invented V2 rocket. 
  • Reply 15 of 20
    XedXed Posts: 2,921member
    DarkMouze said:
    Xed said:
    DarkMouze said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    As noted in the NASA link in the article itself, the first planet directly imaged by a telescope was in 2004, when the VLT telescope pictured one in a system 230 light years away. The mistake referenced in the article was not that they could see a planet using the telescope, but that they were nowhere near being the first to do so.
    I believe VLT (Chile) is the first land-based telescope to take an image of an exoplanet.
    That makes sense to me…I’m betting you’re right. But that’s certainly not what the NASA piece says. I can see the AI headline now. “NASA Completely Botches Astronomy!”
    The link says, "2M1207 b is the first exoplanet directly imaged and the first discovered orbiting a brown dwarf." ¯\_(ツ)_/¯ 
    watto_cobra
  • Reply 16 of 20
    FileMakerFellerFileMakerFeller Posts: 1,561member
    bulk001 said:
    From the headline I thought it got 1+1 wrong. Seems that a product that is in testing is bound to get it wrong. You don’t run headlines like”Appleinsider completely butchered the English language” when a typo slips in. 
    You don't?

    Damn. There goes my idea for a website. :/ 
    watto_cobrabyronl
  • Reply 17 of 20
    XedXed Posts: 2,921member
    urahara said:
    The nearest star is four light years away. There is no way any telescope will be able to see a planet of any star so far away. 
    Are you one of these guys?

    A rocket will never be able to leave the Earth’s atmosphere.” — New York Times, 1936

    This is false news. Rocket has not been invented in 1936. Nazi invented V2 rocket. 
    Dude, you've really got to do a moment of research before posting.

    https://www.nasa.gov/sites/default/files/atoms/files/rockets-guide-20-history.pdf
    watto_cobra
  • Reply 18 of 20
    jellybellyjellybelly Posts: 143member

    This is false news. Rocket has not been invented in 1936. Nazi invented V2 rocket. 
    Um. Remember rockets for warfare and fireworks in China approximately 8 to10 centuries ago in the Sung (née Song as two two English transliteration’s) dynasty?. Vietnam soon after used gunpowder in rockets. Powered by saltpeter plus sulphur as two main ingredients with a few other things thrown in experimentally over time. 
    The V2 was the first successfully controlled liquid fueled rocket, but had less successful predecessors.  
    It’s not helpful to think in binary terms as in false or true, invented or not invented.  
    Developments in science and technology are just that— developments more than inventions.  An even better description is collections of developments of many components and trials of failures that lead to success. 

    The New York Times headline was about people extrapolating for a very long time before 1936 that possibly a rocket could leave the bounds of earth. It was largely a matter of the weight of the fuel and its potential energy density needed to reach escape velocity.  The fuel‘s weight to energy release ratio couldn’t lift its own weight until the ratio improved over time. That eventually developed over decades. But not without unforeseen developments in boosters ( lose the weight by shedding the booster), control, better fuel, dealing with extremely low temperatures and much more.  
    NASA used solid state boosters for a while. Liquid hydrogen was the holy grail as the lightest element with high potential energy when mixed with oxygen, but difficult to manage in a liquid state with needed advancements in refrigeration and insulation and containment. Those pesky tiny atoms are hard to contain.  
    Sidenote: helium balloons deflate overnight because it’s tiny atoms slip through the material of the balloon.  Hydrogen is even smaller and can diffuse through some seemingly solid containers and leak through sealed connections. 

    The V2 was primitive in comparison but a serious (and deadly at its target and as well on the launch sites) development on the long road to flee earth’s gravity. 
    P.S. the Chinese started with rocket assisted arrows.  Cool development but with badly burned archers as well as targets. 
    edited February 2023
  • Reply 19 of 20
    davidwdavidw Posts: 2,120member
     "To err is human, to forgive divine." 

    How can we expect AI to ever be error free (except when solving mathematical equations.), when the algorithms controlling the AI are all written by Humans? And even though we created the AI and are their "God", when the AI makes an error, we tend not to forgive. Once again proving that even as a Creator, we are human.
    gatorguy
  • Reply 20 of 20
    waveparticlewaveparticle Posts: 1,497member
    davidw said:
     "To err is human, to forgive divine." 

    How can we expect AI to ever be error free (except when solving mathematical equations.), when the algorithms controlling the AI are all written by Humans? And even though we created the AI and are their "God", when the AI makes an error, we tend not to forgive. Once again proving that even as a Creator, we are human.
    By constant regression. 
Sign In or Register to comment.