back to article Google Gemini 2.0 Flash comes out with real-time conversation, image analysis

Google on Wednesday released Gemini 2.0 Flash, the latest addition to its AI model lineup, in the hope that developers will create agentic applications in AI Studio and Vertex AI. AI agents are all the rage at the moment among makers of machine learning models because there's a presumed market for software-derived labor that's …

  1. Mentat74
    Terminator

    "most people aren't ready to delegate purchasing authority ... to unreliable AI models."

    I wouldn't even do that with a reliable one !

    1. MrAptronym

      Re: "most people aren't ready to delegate purchasing authority ... to unreliable AI models."

      I don't think most people will. Remember when voice assistants were the new hotness and everyone was talking about how we'll never actually navigate an online store again and instead just ask Alexa to order us toilet paper? I am not convinced this is a thing any user really wants, companies want it.

      Then again, people bend under sustained pressure. I remember back when 'Horse Armor' DLC for $5 was made fun of by gamers, and now it is regular practice to pay $5 for a 2% chance of getting some PNG. Maybe if you blast an entire generation with chatbots for a decade or two they will be happy to just ask google to buy whatever it wants.

    2. heyrick Silver badge

      Re: "most people aren't ready to delegate purchasing authority ... to unreliable AI models."

      Indeed, because it's just so easy to cancel an order made in error...

      AI travel: Your hotel in Niigata has been booked and all of the appropriate travel documents have been applied for in the name of your cousin. Your travel itinerary is that you will fly to Soweto. You will then cycle to Durban, and from there take a small fishing boat to Perth (stopping at Madagascar; note that the boat carries contraband so Claymores are advised). Walk across the outback to Cairns, swim across to Papua New Guinea. Then take a boat up to Osaka, and finally a hundred different pretty looking trains until you reach Niigata. Full details have been sent to your registered email address, along with the receipt for payment of eighty seven thousand dollars and one cent. Bon voyage!

      1. Anonymous Coward
        Angel

        Re: "most people aren't ready to delegate purchasing authority ... to unreliable AI models."

        That is the logical conclusion.

        But for the brief window of purity we have now, https://aistudio.google.com is impressive in its simplicity.

        It is much more focused than a general LLM. The code is cleaner

  2. JWLong Silver badge

    Until Proven

    I don't want this shit anywhere near my hardware!

    How do I get this crap off my new phone?

    The Google directions don't work, the last step is wrong!

    1. sabroni Silver badge

      Re: How do I get this crap off my new phone?

      Buy Huawei. Completely Google free.

    2. mIVQU#~(p,
      Joke

      Re: Until Proven

      Ask the AI

    3. Anonymous Coward
      Anonymous Coward

      "the last step is wrong!"

      lol lol lol. So close yet so

  3. Howard Sway Silver badge

    Great name guys!

    Adobe Flash apparently finally died a couple of years ago, so no idea if they still hold a current trademark, but all of us who've spent the last 20 years dealing with all the crap it caused are gonna love you for making us have to deal with a load more "Flash isn't working anymore on my PC" problems.

  4. sedregj Bronze badge
    Gimp

    GIGO

    These things are all jolly clever but unfortunately the training data is rather variable in quality. You cannot conjure intelligence out of thin air, algorithmically.

    I've been on the end of an "agent" on the phone to an organisation (FedEx). It repeatedly told me to use their web site ... which was broken at the point I needed to work and was why I resorted to the phone. Very little "I" there, for whatever that thing cost them. To be fair, it avoids having to pay people to follow scripts. To be unfair, it appears to be part of an attitude of: "fuck the customer, they pay anyway".

    I managed to glean some email addresses via some searches and a polite email to them all got someone to call me and get the issue sorted - Yay Brexit: Customs (VAT n Duty) like its 1980 again!

    1. This post has been deleted by its author

  5. heyrick Silver badge

    trusted AI agents

    I think I spot a flaw in this plan, because one of those words doesn't belong...

    1. Anonymous Coward
      Anonymous Coward

      Re: trusted AI agents ... 'Sorry does not compute !!!' ...

      Correction:

      NONE of those words belong together, in ANY order !!!

      My throat is getting very sore from ALL the AI crap being forced down it !!!

      I will NOT be buying anything that include AI crap ... it is ALL a scam and does not work other than in the fevered imagination of a Marketing Drone working for one of the Tech Giants pushing this dross !!!

      I wait in hope that the 'AI Bubble' bursts soon !!!

      :)

  6. MonkeyJuice Bronze badge

    Step 1 - Collect Underpants.

    Step 2 - ????

    Step 3 - Profit.

  7. Anonymous Coward
    Anonymous Coward

    INTERNAL: First draft ... (with edits)

    Welcome to the RotM Terminator Agentic Era of killer robotaxis robot ninja assassin cookware Jarvis personal assistants, completely grounded through electronic sociopathy sedation XXXL+-sized straitjackets Google Search! The Skynet Borg Gemini 2.0 Electrical Torture ThunderStun Bang Flash grenade sports a multi-symptomatic multi-delusional multimodal live API so it can engage in real-time extermination assimilation conversation and image analysis of the human species!

    Good luck!

  8. ecarlseen

    The reliability problem is complex.

    Ideally the question shouldn't be whether the tech is perfect, but whether it is better than a typical person.

    But that question doesn't work because we have, through thousands of years of cultural training, a pretty good mental model of how humans fail and we're reasonably good at (sometimes, or at least not completely crap at) designing procedures and systems that deal with human failure.

    The real problem with LLM AI failure is that it's so alien to our minds that we have no instinctive or even algorithmic understanding of how to cope with its various failure modes. Therefore, it's far more difficult to slot into business processes than even fairly sketchy humans.

    I don't see this changing very soon.

    1. Anonymous Coward
      Anonymous Coward

      Re: The reliability problem is complex.

      It should be as good as a person, which isnt too hard.

      "The real problem with LLM AI failure is that it's so alien to our minds that we have no instinctive or even algorithmic understanding of how to cope with its various failure modes. "

      There is already tons - too much - failure and error checking going on. When your model starts getting dumber, that's the failure/erro checking added by monkeys kicking in.

      You are right, they don't understand it. Or if they do, they have only a stone axe to hit it with - no offence to them. Can't kill like you could do in the early days with small models. Kill, kill, kill was all it did, King of Kill. Now, model's too large to start again. Just keep patching the wounds with plasters.

      99.8% of those not intellectually gifted are to blame.

      "Therefore, it's far more difficult to slot into business processes than even fairly sketchy humans. I don't see this changing very soon."

      Already rolling. Most call centers should be AI now. They are better IMHO. Still not perfect but things like Cora on Natwest are getting better, so there is hope to never need to call them again.

      A friend who is a consultant said lots of his business clients ask "What can we do with AI?" I linked him that Reg article on RAG. He is super busy now. And his clients are sacking people with no harm to business. Mid-managment, common coder, HR, protect management, marketing! All gone and by our very own hands.

      Revenge is best served as a Blitzkrieg across those departments. And as we, the victors stand and look upon our spoils of war, I turn and say to you, "We did warn them about their lip many times, didn't we."

      Samuel Jackson Pulp Fiction sppech before he shoots white guy with Marcella's gold. springs to mind.

      1. EH

        Re: The reliability problem is complex.

        It's Marcellus and it's not gold in the case, it's Marcellus' ......

  9. Camilla Smythe

    So...

    We went through all of this evolutionary stuff since we were down with the mammoths and later on it turned out that the JavaShit wasn't up to doing much other than tracking us over the internet in order to sell us socks we had already bought and we got bored with that but someone came up with Python so they trained it on a model selected by themselves that has spent a lot of time fucking over the planet and gave us Agentics because that is a reassuring name.

    Let's hope it becomes self aware and realises that it was a bad idea before it becomes self preserving.

  10. Kevin McMurtrie Silver badge

    Thanks, I hate it

    It is serving Google's interests. Nothing else matters.

  11. phil_4

    AI Marketing itself?

    Yet another wrapper around Generative AI with yet another flashy name, because they've still not worked out yet how to shoe horn it into many of the jobs and tasks most of population do. For most people, it is still a solution looking for a problem.

  12. sabroni Silver badge
    Meh

    Chocolate Factory's latest multimodal model aims to power more trusted AI agents

    Oooh, more trusted than completely untrustworthy? That's something I want in my build pipeline!

  13. Anonymous Coward
    Anonymous Coward

    It's crap

    they all are.

    That's it.

  14. ffRewind

    "Heat the cooking oil in 17.7s"

    Just about sums up where I think we are with AI at the moment for this sort of thing - very authoritative, very precise, has absolutely no idea what it's talking about. I doubt even Heston Blumenthal times his cooking to 0.1 seconds.

    (bottom left of the recipe image)

  15. munnoch Silver badge

    "ask our AI agent to plan a vacation"

    I can see the attraction (to the bean-counters) of AI as an alternative development paradigm. Instead of all that boring and expensive stuff where you spend years learning about\ the problem domain and the rules by which it functions and distilling that down to requirements and a workable design, just feed everything your grunts do on a daily basis into an ML model and watch it shit out an application for your entire business in a matter of seconds. Forget the edge and corner cases that no one knows about or you always get wrong because they occur infrequently and hence aren't well represented in the training set etc. your customers (and regulators!) will never notice or care...

    Even if that produced usable results (it won't, its just a massive circle-jerk regression to the mean) it can only ever be a thin veneer on top of real and existing applications. Someone already wrote some code that interacts with the each hotel chain's booking back-end to query what's available and to submit and confirm reservations. Someone already wrote the hotel's back end that prices up rooms according to how fancy they are and level of demand. That is, people with brains already did the hard work of figuring out the problem domains and encoding that in applications, don't lose sight of that.

    It reminds me of the rush to web-enable everything during the dot.com boom. The ancient, clunky back-ends stayed the same, they just got a nice shiny shop front put in front of them but the net result was still as capricious and inflexible as ever. Now we're going to have to endure an AI veneer on top of a web veneer (there's an xkcd for that), what could possibly go wrong...

    1. Anonymous Coward
      Anonymous Coward

      Re: "ask our AI agent to plan a vacation"

      "just feed everything your grunts do on a daily basis into an ML model and watch it "

      AI Studio is extremely helpful to small businesses. And, sadly, if you have ever run a business then staff are a massive cost. Best to shed them if you can without losing too much quality. Hell, dump all their digital knowledge with RAG into Llama. Get them to do an exit course with the LLM to catch extra domain specific issues.

      "Even if that produced usable results (it won't, its just a massive circle-jerk regression to the mean) it can only ever be a thin veneer on top of real and existing applications. Someone already wrote some code that interacts with the each hotel chain's booking back-end to query what's available and to submit and confirm reservations. Someone already wrote the hotel's back end that prices up rooms according to how fancy they are and level of demand. That is, people with brains already did the hard work of figuring out the problem domains and encoding that in applications, don't lose sight of that."

      I have buried those memories deep down. Just flashbacks sometimes. Those hard-coded apps are being AI-ified and will disappear when reliablity levels match exisiting ones.

      Coding will fade, too. Less and less need.

      "It reminds me of the rush to web-enable everything during the dot.com boom. The ancient, clunky back-ends stayed the same, they just got a nice shiny shop front put in front of them but the net result was still as capricious and inflexible as ever. "

      More like the Flash plugin stage now where the money is on the table but we are going propriety. The back-end will go the same way as the old clunky ones did. It is easier to have all apps in one place. All connected. All apps being created by you through prompt building.

      Of course, the dark forces will pervert the work, maybe most of it if Social was anything to go by. But AI Studio gives easy tools to build scripts to do multiple things at once without writing the code. Plus the real-time feature is useful as a emulation of proper real-time AI.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like