News: 1715773513

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Google thinks AI can Google better than you can

(2024/05/15)


In the future, "Google will do the Googling for you" – or so suggests Liz Reid, VP of search.

On Tuesday at Google I/O, the search ad firm's annual developer conference, executives made the case for a world where multimodal machine learning models connect the dots and fill in the blanks.

"With AI Overviews, Google does the work for you," [1]said Reid. "Instead of piecing together all the information yourself, you can ask your question."

[2]

[3]AI Overviews – offered previously as a Search Labs experiment – is rolling out to US search users today, with more countries to follow.

[4]

[5]

Google's vision for the future of search involves posing complex queries that combine multiple demands into a single directive.

For example: "Find the best yoga or Pilates studios in Boston and show me details on their intro offers, and walking time from Beacon Hill." Normally this would be three or more keyword search queries. But by plugging Gemini into the search process, the entire inquiry can be handled at once.

[6]

"Under the hood, our custom Gemini model acts as your AI agent using what we call multi-step reasoning," explained Reid. "It breaks your bigger question down into all its parts. And it figures out which problems it needs to solve and in what order."

The presentation of AI Overview results, however, made it look like Google's search page will keep much of the viewer's attention, rather than sending searchers to linked websites.

According to Reid, Gemini can also assemble a multi-day meal plan on demand or handle activity planning – at least for those happy to delegate decisions to an AI model. Meal and trip plans, available in Search Labs, can be exported to Gmail or Docs from icons on the search results page, and additional plan categories – like parties, date nights, and workouts – are being developed.

[7]

Another pending search capability, coming soon to [8]Search Labs , is the ability to search using videos as input.

[9]

Using 21st century tools to explain technology from the 19th – click to enlarge

In an onstage demo, Rose Yao, VP of product at Google Search, showed how a video of a turntable with a moving tonearm could be submitted to Google Search to answer the question "Why will this not stay in place?" The underlying Gemini model is savvy enough to understand that "this" refers to the wobbly tonearm and to return a recommendation about how to fix the issue.

Astra, your AI pal who's fun to be with

Demis Hassabis, CEO of Google's DeepMind, took a turn on stage to tease [10]Project Astra – a "universal AI agent."

"For a long time, we've wanted to build a universal AI agent that can be truly helpful in everyday life," enthused Hassabis.

"Building on our Gemini model, we've developed agents that can process information faster by continuously encoding video frames, combining the video and speech input into a timeline of events, and caching this for efficient recall. We've also enhanced how they sound, with a wider range of intonations.

"These agents better understand the context you're in, and can respond quickly in conversation, making the pace and quality of interaction feel much more natural."

A [11]prerecorded video – said to have been captured in a single take in real time – showed off the Chocolate Factory's progress toward this goal. It featured an employee walking around a Google office scanning it with a smartphone running Astra.

Asked to report when it sees something that makes sound, the agent on the phone responded in a human-sounding voice that it spotted an audio speaker. It then went on to identify the speaker's tweeter after that portion of the speaker had been pointed to in the phone's live video of the room.

[12]

Astra knows its woofer from its tweeter – click to enlarge

Thereafter, the agent composed an alliterative phrase to describe a bowl of crayons upon request. And asked to explain source code on a nearby monitor, the software helper replied that it defines encryption and decryption functions.

Finally, in response to a query about its location, the agent determined that it was in the King's Cross area of London – apparently based on the view from the office window. It also remembered where the user had left their glasses – which many users might find worth the price of admission on its own.

According to Google, this is an ongoing project and some of these capabilities can be expected in Gemini at some point in the future.

[13]OpenAI says natively multimodal GPT-4o eats text, visuals, sound – and emits the same

[14]Google says its 'Trillium' TPUs are ready to power the next-generation of AI models

[15]Arm to launch AI division with new chips by 2025

[16]AWS CISO tells The Reg: In the AI gold rush, folks are forgetting application security

Gemini furthered its [17]infiltration of Google Workspace , with Gemini 1.5 Pro now providing AI help for Workspace Labs and Gemini for Workspace Alpha users via the side panel of Gmail, Docs, Drive, Slides, and Sheets. It will be available next month to [18]Google One AI Premium subscribers on desktop via Gemini for Workspace add-ons.

Gemini Workspace features are also coming soon to the Gmail mobile app – specifically email summarization – in June, with contextual smart reply and Gmail Q&A out by July.

Creatives, meet your doom

Not to be outdone by OpenAI's text-to-video model Sora or its text-to-image model DALL·E, Google [19]announced [20]Veo and [21]Imagen 3 for creating videos and images from text prompts.

"Veo creates high-quality, 1080p videos from text, image, and video prompts," declared Hassabis. "It can capture the details of your instructions in different visual and cinematic styles. You can prompt for things like aerial shots of a landscape or a time lapse, and further edit your videos using additional prompts."

Imagen 3 looks interesting for its ability to generate legible text within images and its ability to edit areas of an image by selecting and re-describing what should appear in the selection area.

Veo is available through a private [22]waitlist via Google's [23]VideoFX service, while Imagen 3 is [24]similarly gated via [25]ImageFX . [26]MusicFX and [27]TextFX are already available.

With these new AI capabilities, Google is also expanding its AI safety mechanisms. Specifically, the Chocolate Factory is making its SynthID watermarking technology [28]available for Gemini-generated text in the app and on the web and for Veo video.

For more on Google's AI model enhancements, see our report on the megacorp's [29]developer-oriented announcements. ®

Get our [30]Tech Resources



[1] https://youtu.be/XEzRZ35urlk?list=PL590L5WQmH8doPo8OufXavO2Qu4ysZjyl&t=7142

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZkTcIeXUke1LxC6MDK4sEwAAAMw&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[3] https://blog.google/products/search/generative-ai-google-search-may-2024/

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZkTcIeXUke1LxC6MDK4sEwAAAMw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZkTcIeXUke1LxC6MDK4sEwAAAMw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZkTcIeXUke1LxC6MDK4sEwAAAMw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZkTcIeXUke1LxC6MDK4sEwAAAMw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[8] https://labs.google/

[9] https://regmedia.co.uk/2024/05/14/google.jpg

[10] https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/#ai-agents

[11] https://www.youtube.com/watch?v=nXVvvRhiGjI

[12] https://regmedia.co.uk/2024/05/14/speaker1.jpg

[13] https://www.theregister.com/2024/05/13/openai_gpt4o/

[14] https://www.theregister.com/2024/05/14/google_tpu_trillium/

[15] https://www.theregister.com/2024/05/13/apac_roundup_may13/

[16] https://www.theregister.com/2024/05/13/aws_ciso_ai_security/

[17] https://blog.google/products/workspace/google-gemini-workspace-may-2024-updates/

[18] https://one.google.com/about/plans?utm_source=g1&utm_medium=web&utm_campaign=gemini_announce

[19] https://blog.google/technology/ai/google-generative-ai-veo-imagen-3/

[20] http://deepmind.google/technologies/veo

[21] http://deepmind.google/technologies/imagen-3

[22] http://labs.google/VideoFX

[23] https://aitestkitchen.withgoogle.com/tools/video-fx

[24] http://labs.google/trustedtester

[25] https://aitestkitchen.withgoogle.com/tools/image-fx

[26] https://aitestkitchen.withgoogle.com/tools/music-fx

[27] https://aitestkitchen.withgoogle.com/tools/text-fx

[28] https://deepmind.google/discover/blog/watermarking-ai-generated-text-and-video-with-synthid/

[29] https://www.theregister.com/2024/05/14/google_io_is_google_ai/

[30] https://whitepapers.theregister.com/



Big Deal

boblongii

Given that Google already ignores most of what I ask for and simply throws up a Wikipedia page I imagine a brain-damaged pigeon could "do my Googling for me" with similar results.

Google are in complete denial about how shit their service has become.

I don't care

cschneid

> Google thinks AI can Google better than you can

I'm certain Google thinks that, but Google is wrong.

Re: I don't care

Neil Barnes

But, but, but... Google _knows_ that you want lots of adverts for products completely unrelated to your search. It's just that you haven't realised it yet.

Google missteps

Reiki Shangle

Oh jeezuz.

We have Bernie-Lee to thank for these glorified, excessive infostructure systems.

Now, can AI help me fill this form for my Rail Pass renewal whilst the Javascript assumes I want to move automatically to the next field of my Passport numbers? Can the AI take over the UI testing and forward the expletive-filled results to the corresponding AI dev?

Thanks. Thanks a lot.

Hidden variables revealed

SW10

"Find the best [has paid most for ads] yoga or Pilates studios in Boston and show me details on their [and high-paying advertisers in Boston] intro offers [exclusive to Google], and walking time [plus paid Uber ad] from Beacon Hill [showing ads for attractions in Beacon Hill]."

Re: Hidden variables revealed

tiggity

@SW10

Indeed - what is "best" has lots of interpretations based on the user, generally unrelated to how much cash Google make from the studio / gym..

e.g. as outfits used for yoga / Pilates can be quite clingy & revealing, then can be a bit embarrassing for some women to do yoga in a mixed class (unless they switch to full on very baggy outfits), especially with some of the poses.

So if a woman is uncomfortable with male behind gazing at her perineum in some poses then she would be likely to rank a studio higher if they offered women only classes.

Whereas, as a bloke I don't care, so single or mixed sex would be irrelevant to me: Though a pervy male* might be very happy with the idea of a mixed sex class so that would be "best" for them.

* My (F) partner told me about an old bloke (he would probably not have been able to manage the actual class as quite a physically demanding one) who used to really obviously letch at women doing dance fit class she attended (quite a lot of bouncing bosoms even with people wearing sports bras) as that gym area was visible from a public viewing area where people could sit & have coffee etc. So blokes going to a leisure centre / gym to perv is not unknown. In this case old bloke was eventually barred after complaints from users of various classes (he would probably have got away with it had he been less obvious about it).

We are well aware

Omnipresent

of how much damage robots of death controlled by pure evil can do given the chance.

Google and micro$oft are about to go the way of the dodo bird. If apple isn't very careful, they will go down with them.

Re: We are well aware

DJO

Dream on, people have been saying IBM are on their last legs for over 40 years but they keep slogging on, the same is and will continue to be true for Google, Microsoft & Apple.

(Many companies should probably be consigned to the dustbin of history, but for these biggies it ain't happening soon - Some will, Tesla has 3 or so years before it is either bankrupted or sold to some adults).

Re: We are well aware

Omnipresent

how strange you don't remember the world before microsoft and google. It wasn't so long ago.

Micro$oft has been labelled a national security risk, and google's next step is to NOT let you search for information.

Apple has aligned it's self with google, and to be totally honest with you, I no longer trust them. I can't be the only one who feels this way. Much like social media it's self, The tech may never go away, but the companies do.

All it will take is someone with another option.

MrAptronym

If google is simply going to be scraping pages and giving you their information without even linking through to the actual websites then they are just a parasite on the people and websites who actually make content.

For years they have been trying to keep your traffic on google, or on the pages that pay for ads: now they can simply not link anything and give you scraped answers.

Doctor Syntax

How about a prompt along the lines of "List links to pages that show ... and leave out all the crap"

AI Search is shit

I ain't Spartacus

Google must have AI checking their customer satisfaction or something.

The more they put "AI" into their search, the shittier their search becomes. Google is now filled with these little paragraphs of, usually unattributed, snippets (or machine learning summaries) from various websites as well as "other people also asked this entirely not similar question" - which you can click on to get some even less relevant shit results.

A couple of times I've caught the ML summaries being totally wrong as well.

The other day I looked up the number of days in 80-odd years, because my nephew wanted to know how many days his Nan (my Mum) had been alive. It came out with 230,000-odd - which is wrong by an order of magnitude. So a striaght mathematical question, how many days between these two dates. That search also didn't throw up one of the many websites that exist which will easily do this calculation for you, but a quick new search for one of those got it for me. Google being shit meant it would almost have been quicker to work it out myself.

OK, tirivial example. But that should be a really easy query for a language engine to parse, and then a really easy anwser to spit out. I've also done searches on historical characters, to get summary paragraphs that are complete bollocks. Worse because they're using ML to summarise stuff, almost the only source they actually admit to is Wikipedia, so except for those attributed paragraphs the rest of it has to be regarded as junk.

To be fair, the Wiki summaries I've seen have been reasonable one paragraph summaries when I've clicked on the aritcles to check.

But they're seemingly doing no testing to see if the stuff they're using now works. And rather than working on that, they're going to try to add complexity so the results are now even more unreliable?

What a shitshow.

Excuse me while I end this post here, I just need to search Google for "alternative search engines that aren't as shit as Google"...

Re: AI Search is shit

MrAptronym

That is such a huge and obvious problem, and AI slinging companies keep just ignoring how often these models are wrong. The feeling really seems to be that magically these models will stop hallucinating because... technology always gets better? If the entire point is that these AI searches are supposed to be saving you time by compiling information then they need to be really consistently correct. If I need to verify any AI answer like that to make a decision, then what are we even doing here? Even if it was right 99% of the time, it would still be too unreliable for anything meaningful.

For the questions they give as examples it is even more unbelievable: What does it mean to be the best yoga studio in Boston? How would determine that? That is a case where personal preference is obviously important. Optimal cost, exercise intensity and atmosphere will clearly be different for different people. AI companies keep using these examples of things like trip planning or finding a service business, but those are among the last things I would trust to a system like this. This feels just like the rush for voice assistants a decade ago: people may use them to get the weather, control lights or play music but almost no one trusted them to even buy toilet paper for them.

I’ll believe it when I see it

I am David Jones

Wasn’t it google that faked the phone call of a digital assistant booking a restaurant?

Google search cannot be trusted

Marty McFly

Very simple case study during Covid...

A search for "Covid vaccination card" produced no results on Google. Same search on other engines (duckduckgo, startpage, etc) provided links to *government* websites with blank vaccination cards, in addition to lots of other public-facing documentation from the CDC. Essentially 100+ page documents on how a state is responding to the pandemic and buried in the middle was a blank vaccination form.

This was proof that Google will manipulate search results to what they want me to see and not the results that I am seeking.

It is not Google's responsibility to police legitimate content. Again, these were government produced documents, hosted on government websites (ie: State of xxx), that Google was intentionally concealing from the public during a time of national crisis. These were not shady backwater websites laden with malware that Google was hiding.

In a lot of ways the only value Google has is the user's trust and faith in the validity of their search results. I stopped using Google search when I realized they were intentionally hiding valid results from being presented in their algorithm. They will never regain my trust.

Article photo: "Google will do the Googling for you"

Dan 55

Nooo, [1]you can break the Internet .

[1] https://www.youtube.com/watch?v=Neikj_kcbOI

History repeats itself. That's one thing wrong with history.