[HN Gopher] Gemini Live with camera and screen sharing capabilities
       ___________________________________________________________________
        
       Gemini Live with camera and screen sharing capabilities
        
       Author : agnosticmantis
       Score  : 185 points
       Date   : 2025-04-07 16:53 UTC (4 days ago)
        
 (HTM) web link (blog.google)
 (TXT) w3m dump (blog.google)
        
       | polishdude20 wrote:
       | Why only Pixel 9? Surely none of the computation is on-device
       | anyways
        
         | ipsum2 wrote:
         | The second sentence contradicts your statement
         | 
         | This soon will be available to all Gemini Advanced subscribers
         | on Android devices, and today we're bringing it to more people,
         | starting with all Gemini app users on Pixel 9 and Samsung
         | Galaxy S25 devices
        
           | bytesandbits wrote:
           | curious, what is done on-device?
        
             | echelon wrote:
             | Likely nothing. Mobile devices can't do much AI stuff
             | except for the tiniest of models, and it'll likely be a
             | long time before they will be able to anything super
             | useful.
             | 
             | Edge compute is a long ways off, even on desktop without a
             | dedicated GPU. But especially on mobile.
             | 
             | You might have bumped into a media website trying to run a
             | WASM-powered onnx runtime background removal tool, or
             | perhaps a super slim LLM. You'll notice how slow these are
             | and how they can lock up your browser. That's about the
             | experience you can expect from edge compute.
             | 
             | Nvidia's proclamation that they're going to be working on
             | robotics as their next growth sector could mean more
             | innovation on the edge / low power compute front. But most
             | of the yield will come from better model architectures and
             | models designed specifically to work with compute
             | constraints.
             | 
             | For now, datacenter inference reigns supreme.
        
           | addandsubtract wrote:
           | I still can't use the "summarize" feature that was used in
           | marketing to sell the Pixel 8 Pro. Not holding my breath on
           | ever being able to use this, either.
        
         | cavisne wrote:
         | You can do the same from the browser on any phone (including
         | iphone) by going through AI Studio
         | 
         | https://aistudio.google.com/live
        
           | clueless wrote:
           | Upvoted. For a couple of weeks now, I added that site as a
           | bookmarked app on my iPhone just to use that Gemini live
           | feature, though I can only use the camera/voice feature (no
           | screen sharing ability on the chrome ios browser afaik).
           | Super weird they are not releasing this feature more
           | officially on iOS
        
         | colecut wrote:
         | "The Google Pixel 9 is powered by the Tensor G4 chip, a custom-
         | designed processor optimized for AI. This chip, along with 12
         | GB of RAM (or 16 GB for Pro models), enables advanced AI
         | features and enhances overall performance."
        
           | cavisne wrote:
           | None of that onboard stuff ended up working afaik as the
           | Tensor chips shipped right as giant LLM's became how
           | dominant.
        
         | 0x1ceb00da wrote:
         | Incentives to buy new hardware.
        
       | sureglymop wrote:
       | Watching that demo video, I wonder why they chose to use that?
       | 
       | Gemini only talked about some useless surface knowledge that
       | would be forgotten quickly, whereas if she actually read the
       | Wikipedia page she would learn more and retain it better.
        
         | clueless wrote:
         | right? not great examples. Here are the kind of prompt examples
         | I would have used instead:
         | 
         | - explain this symbol to me
         | 
         | - what TV show am I watching
         | 
         | - how much is the house I'm looking at worth (bring in geo-
         | location)
         | 
         | - How often should I be watering this plant
         | 
         | - how many calories are in this meal
         | 
         | - what type of car is this
         | 
         | - whats the exact name/type of this screw part
         | 
         | - what are the exact dimensions of this door frame
         | 
         | - what kind of style of art is this
         | 
         | - how often should I be taking this medication
         | 
         | - how many copies has this book sold
         | 
         | - which part of the world is this pic from
        
           | dimitri-vs wrote:
           | Because besides the TV show one none of your examples benefit
           | from the video/"live" modality? Or at least I wouldn't be
           | motivated to use video for that.
        
           | com2kid wrote:
           | My meta glasses can do a decent number of those, with the
           | ideal form factor for those use cases.
        
             | chem83 wrote:
             | Looks like Google was demonstrating some of that on stage
             | recently: https://gizmodo.com/google-showed-off-advanced-
             | ar-glasses-th...
        
           | kristopolous wrote:
           | or even the internet one. It would have been nice if it was
           | like "I see you have a NetMan 402-BX with a blinking amber
           | light. That's a common failure mode that can be fixed by
           | updating the firmware. You need to press and hold the power
           | button and put a pin into the reset hole..."
           | 
           | You know ... something more than "reboot that shit"
        
           | fc417fc802 wrote:
           | > how often should I be taking this medication
           | 
           | What could go wrong?
        
         | bn-l wrote:
         | It's aimed at normies. What you watched would blow their minds
         | and be exactly what they were hoping for.
        
       | tintor wrote:
       | "Screen images simulated."
        
       | sksxihve wrote:
       | [flagged]
        
         | dimitri-vs wrote:
         | Don't be so pessimistic, how about these uses:
         | 
         | - Let Gemini watch you drive so you can get advice on driving
         | better
         | 
         | - Let Gemini help you rewire your outlet for the first time
         | 
         | - Let Gemini help you pick edible mushrooms in the forest
        
           | jmugan wrote:
           | It's funny, you are kidding, with the impending death and
           | such, but having it be able to look at what you are seeing
           | seems useful to me. E.g., "How do I work this contraption?"
        
           | devsda wrote:
           | > Don't be so pessimistic, how about these uses:
           | 
           | I suspect, these examples are chosen specifically because
           | their failures modes can be lethal ?
        
             | kibibu wrote:
             | How about:
             | 
             | - Gemini, help me tell if this is sarcasm
        
           | qingcharles wrote:
           | "Gemini, what's the best way to bury this body?"
           | "Gemini, please count the cockroaches in this kitchen."
           | "Gemini, calculate my survival chances sprinting across this
           | interstate."       "Gemini, speculate on the percentage
           | volume of pee-pee in this swimming pool."       "Gemini, how
           | many days will this meal set back my life expectancy?"
           | "Gemini, will these socks lead to my arrest by the fashion
           | police?"       "Gemini, my cat is making that face again --
           | has it found out it's adopted?"
        
           | terminatornet wrote:
           | About those edible mushrooms -
           | https://www.vox.com/24141648/ai-ebook-grift-mushroom-
           | foragin...
        
         | dullcrisp wrote:
         | It seems that there's a subset of people who will agree
         | unironically.
        
           | sksxihve wrote:
           | They are the target market for this
        
         | meta_ai_x wrote:
         | I'm sure you laughed at the words and sentences GPT-2
         | generated.
        
         | thorum wrote:
         | They're listing the use cases where a 10-20% hallucination rate
         | isn't a big deal. They can't advertise the truly useful
         | applications of the tech because it's not reliable enough
         | (yet).
         | 
         | But they can make it more reliable over time by deploying the
         | product now and training on user data:
         | 
         | > Google collects your chats (including recordings of your
         | Gemini Live interactions), what you share with Gemini Apps
         | (like files, images, and screens), related product usage
         | information, your feedback, and info about your location. (...)
         | 
         | > Google uses this data, consistent with our Privacy Policy, to
         | provide, improve, and develop Google products and services and
         | machine-learning technologies, including Google's enterprise
         | products such as Google Cloud.
         | 
         | https://support.google.com/gemini/answer/13594961?hl=en
        
           | sksxihve wrote:
           | Google has a history of faking their Gemini demos, even if
           | this demo was impressive (it's not) I would feel the same way
           | about it. Yeah, it will get better in time like self driving
           | cars, quantum computers, super conductors, nuclear fusion.
        
         | dang wrote:
         | " _Don 't be snarky._"
         | 
         | " _Please don 't post shallow dismissals, especially of other
         | people's work. A good critical comment teaches us something._"
         | 
         | https://news.ycombinator.com/newsguidelines.html
        
       | vorpalhex wrote:
       | Is anyone seriously using Gemini daily? How is it compared to
       | other agents you've tried? Do you feel it's a good value prop?
       | What does it excel/fail at?
        
         | FactolSarin wrote:
         | I already pay for Google One storage, so it's the cheapest of
         | the paid LLMs for me. That's 99% of the reason I use it, and
         | honestly I don't really have any strong opinions on it compared
         | to ChatGPT. It's about the same level, and with new models
         | constantly being released for all the different LLMs I've kind
         | of lost track of what it's particularly good or bad at compared
         | to others.
         | 
         | I will say the video with Gemini live is pretty impressive. My
         | family and I tried it a bit yesterday, and my kids wanted to
         | show Gemini all our pets. My kid showed it our cat, picking it
         | up roughly as she is wont to do, and I was impressed when it
         | asked "Is [name of cat] always so patient being handled like
         | that?"
        
           | diggan wrote:
           | > I already pay for Google One storage, so it's the cheapest
           | of the paid LLMs for me. That's 99% of the reason I use it,
           | and honestly I don't really have any strong opinions on it
           | compared to ChatGPT
           | 
           | I'm on a ChatGPT pro plan, been using it for a good while but
           | got an offer on Google One storage so tried it out for a
           | month. Google's models are far behind compared to OpenAI's,
           | and seemingly o1 Pro Mode is still the best out there, albeit
           | slow obviously. But probably the model I've got furthest with
           | on difficult problems, and even the "simpler" models from
           | OpenAI are still better than Gemma 2.5.
           | 
           | It does seem that Google has better tooling available for
           | their models though, so a combination of the tooling of
           | Google with the models of OpenAI would probably be optimal,
           | but unlikely we'll see that happen.
        
             | whimsicalism wrote:
             | how long ago? i think 2.5 exp is better than o1 or o3-mini
             | in my experience.
        
         | lytedev wrote:
         | I'm probably on the wrong side of history and generally think
         | LLMs are way overhyped, but Gemini 2.5 Pro is the first one to
         | consistently impress me for pretty much everything.
        
         | Spooky23 wrote:
         | I use it and ChatGPT. I think they are pretty close. Gemini has
         | gotten a lot better, and I think it's superior on the deep
         | research side. NotebookLM is incredible.
         | 
         | For more general things, ChatGPT is still better imo, but that
         | gap is shrinking, at least for what I do. Their analysis plugin
         | is super useful to me.
         | 
         | I will say that it's hard to evaluate as some of my habits may
         | drive behavior.
        
         | makeitdouble wrote:
         | Not daily, but a decent amount I think.
         | 
         | Image search and text recognition have become way better than
         | Google Lens ever was. In-place translation is also nice, and I
         | use to check idioms, and other grammarly-esque things.
         | 
         | I think it's good at language related things and
         | transformations, not great at anything else (I wouldn't ask it
         | to create anything new TBH). I can't even get it reliably spun
         | Google docs from my prompts from the standard web/app
         | interfaces.
         | 
         | Similar to FactolSarin, it's baked into my Google account and
         | subscriptions so I don't really need it to be stellar and
         | wouldn't pay for it as a single product.
        
         | Workaccount2 wrote:
         | Up unit 2.5 it was kinda meh, but now it's my daily driver.
         | Still have ChatGPT, but swapped out Anthropic.
        
         | mvdtnz wrote:
         | You can't upload code files (.java files, build.gradle files,
         | etc) on the web UI, which makes it 100% worthless for my dev
         | flow. I'm not interested in integrating it with my IDE or
         | installing CLI tools, I want to just upload relevant code and
         | get results.
        
           | gundmc wrote:
           | Try using the interface via ai.dev instead of
           | gemini.google.com. I upload code files all the time and it
           | works.
        
         | phonon wrote:
         | Gemini 2.5 Pro is very competitive with o1/o1-pro in reasoning
         | and has a larger maximum context. It also just rolled out Deep
         | Research which is also very competitive (perhaps even better)
         | than Open AI's.
        
           | bearjaws wrote:
           | I'd argue that 2.5 pro is better because it doesn't suffer
           | from GPT-isms, where its blatantly obvious someone used
           | ChatGPT.
        
             | CamperBob2 wrote:
             | It's also much, much faster than o1-pro. That feels like a
             | bigger win every time I use it.
        
         | kristopolous wrote:
         | I use it in coding. It's pretty good.
        
         | theblazehen wrote:
         | I find it's a great complement to Sonnet 3.7 in agentic
         | programming workflows, the extended context is amazing, and it
         | can often solve issues that Sonnet 3.7 can't handle (and vice-
         | versa)
        
         | PunchTornado wrote:
         | I use it daily, even started planning trips, explaining stuff,
         | coding. Easily the best ai out there, then claude and then
         | deepseek.
        
       | kovek wrote:
       | I think it would be nice if the Pixel Fold could do: Have a
       | browser on the left showing some content, and have Gemini on the
       | right, where you can prompt it with questions or asking it to
       | take actions on the left.
        
       | discordance wrote:
       | and then humanity became illiterate
        
       | behnamoh wrote:
       | This is Apple Intelligence the way it was supposed to be ("AI for
       | the rest of us"), but Apple just doesn't "get" AI, so here we are
       | --the only platform provider that is taking the correct approach
       | to AI is Google.
        
         | Workaccount2 wrote:
         | It seems very likley that Apple will license this in some form.
         | They already are testing with ChatGPT and Gemini, but I don't
         | think OpenAI is going to be able to undercut Google on price.
         | And how it looks right now they might not be able to beat them
         | on performance either.
         | 
         | Deep down though I can feel Apple must have some level of
         | aversion to this. They have been trying to untangle themselves
         | from Google for so long.
        
       | cadamsdotcom wrote:
       | Marketing should use their imagination!
       | 
       | Imagine putting dice and random objects (cups, forks..) on a
       | table, pointing your phone at them and asking it to invent a new
       | game for your friends. Tell it to use these objects and also use
       | the live camera as a gameplay element.
       | 
       | Or recognizing bird or plant species.
       | 
       | Or helping a blind person go hiking, helping avoid tree roots and
       | describing the beautiful scenes they're in.
       | 
       | So much possibility!
        
         | wgd wrote:
         | How charitable of you to assume those examples work reliably.
        
           | jryle70 wrote:
           | Are you willing to bet that it wouldn't work reliably in a
           | year, 2 years, 5 years?
        
             | pfg_ wrote:
             | If you're releasing something today, should you talk about
             | what it can do now or what it might be able to do in two
             | years?
        
           | jskarpet wrote:
           | Bemyeyes app already work quite reliability to describe
           | scenes to the blind
        
             | lynx97 wrote:
             | Haha, did you evaluate this personally?
             | 
             | I did a BeMyEyes test recently, trying to sort about 40
             | cans according to the existance of a deposit logo. After 90
             | minutes of submitting photos, and a second round to make
             | sure it doesn't lie too much, I had 16 cans which according
             | to BeMyEyes (OpenAI) had a deposit logo. Then, I went to
             | the shop to bring them back. Turns out, only _4_ cans had a
             | logo. So after a _second round_ to eliminate
             | hallucinations, the success rate was only 25%.
             | 
             | Do you call that _reliable_?
        
               | diggan wrote:
               | > I did a BeMyEyes test recently
               | 
               | But isn't the BeMyEyes assisting happening via other
               | humans? I remember signing up for some "when blind people
               | need your help" thing via BeMyEyes and I understood it as
               | it's 100% humans on the other end of the call that will
               | help you.
        
               | woadwarrior01 wrote:
               | That changed a while ago. They also use OpenAI's APIs
               | now.
               | 
               | https://openai.com/index/be-my-eyes/
        
               | lynx97 wrote:
               | Yes, what you are describing is how BeMyEyes started, and
               | it still offer that feature.
               | 
               | However, somewhere around 1 or 2 years ago, they added a
               | OpenAI vision model based way to send in photos and have
               | them described.
               | 
               | In general, its a very nice feature, if it works. For
               | instance, I do use it successfully to sort laundry.
               | 
               | But the deposit logo test I did gave horrible results...
        
         | notachatbot123 wrote:
         | > Or helping a blind person go hiking, helping avoid tree roots
         | and describing the beautiful scenes they're in.
         | 
         | "If only you could see, then you could appreciate the beauty of
         | the scenery! Let me describe what are missing out on. I hope
         | your sub-standard experience, as rated by sighted people, is
         | sufficiently adequate to you. May I list more short-comings of
         | your existence in comparison to able people?"
        
         | lynx97 wrote:
         | Blind person here. I can confidently say that vision AIs for
         | navigating the real world are pretty much hype. It is a
         | bandwidth and latency issue. Latency from camera to AI, and
         | bandwidth from device to brain... Please, don't hype
         | accessibility just for your personal fun. There are people out
         | there with real problems, and dangling impossible solutions
         | before their "eyes" is pretty much cruel.
         | 
         | > describing the beautiful scenes they're in
         | 
         | And this part is ableism at its best. Do you eally think what
         | people like me are missing are computer generated descriptions
         | of our surroundings, so that we notice the beauty? Reminds me
         | of the Cochlear implant debate some people from the deaf
         | community have. Is my life less interesting because I dont see
         | _your_ colours? Am I a lesser human that needs treatmenet
         | because I dont see _your_ beauty? Me thinks no.
        
           | JohnPrine wrote:
           | Agreed, blind people and sighted people are able to
           | experience different kinds of beauty that are inaccessible to
           | each other. As a sighted person, if there was a technology
           | that could describe the beauty that a blind person
           | experiences in a given moment I would think that would be
           | pretty cool to try.
           | 
           | > Please, don't hype accessibility just for your personal
           | fun. There are people out there with real problems, and
           | dangling impossible solutions before their "eyes" is pretty
           | much cruel
           | 
           | Do you think it's going to be impossible forever? bandwidth
           | and latency seem like the surest things to improve in AI tech
        
             | lynx97 wrote:
             | I don't know about bandwidth from device to brain. Maybe
             | Neuralink or something similar?
             | 
             | It has been tried a lot. I saw the first sonar-alike
             | navigation aid in the early 90s. It basically translated
             | detected obstacles into vibration. Thats where you start to
             | realize that bandwidth is the issue. Because a single, or
             | maybe even a group of, vibrations, doesn't really tell you
             | anything about the nature of the obstacle. Now, we're at a
             | point where vision models (if they dont hallucinate,
             | DANGER!!!) can actually destinguish different obstacles and
             | even describe them. Nice. However, you're pretty much
             | limited to speech synthesis as an information channel. That
             | is not _that_ bad, but very distracting when you move about
             | outside. After all, blind people already use their ears to
             | learn about their surroundings. There isn 't much, if any,
             | bandwidth left to stick some constantly chatting voice in.
             | You end up deminishing your actual senses input about the
             | surroundings, which is also dangerous. Nothing beats the
             | turn-around time of your own ears or tactile info you might
             | get from your cane...
             | 
             | So, to answer your question: Maybe. I haven't seen a
             | technology that can squeeze some more bandwidth out of the
             | current situation yet.
        
               | JohnPrine wrote:
               | Thanks for the insight. Very cool to get your perspective
               | on this
        
       | randomsofr wrote:
       | pretty useless demos haha, i wonder why they choose those cases,
       | maybe it really doesn't do much else correctly rn
        
       | sandspar wrote:
       | Cool! I'm looking forward to having a live AI drawing tutor. None
       | of the models are there yet but we're getting close!
        
       | devinprater wrote:
       | It's not the best at helping me play video games yet, lol. Ah
       | well. Blind people are used to waiting. :)
        
       | simonw wrote:
       | I've been using this to help me read papers with mathematical
       | notation in them. I screen share with
       | https://aistudio.google.com/live and then ask (out loud)
       | questions like "what do these symbols mean?" - it's a huge help
       | for me, since I never memorized all of that stuff.
        
         | rockmeamedee wrote:
         | Oh this is great! I always have this problem. I find that's one
         | of my biggest barriers when reading queueing theory content.
         | I'm only doing it intermittently so I don't have memorized the
         | meanings of r,s,m,l...
         | 
         | Visually I also often confuse rho and sigma, and math texts
         | will use psi ps and phi ph in weird fonts and I can never tell
         | them apart.
        
       | yalogin wrote:
       | I am getting extremely skeptical of AI "age". I was hoping it
       | will unlock a whole new generation of startups like the iPhone
       | did. However genAI is too generic and too blunt a tool, as in it
       | does everything. However it's too expensive for a small company
       | to do stuff with. Looks like these AI companies (google and
       | OpenAI) realize that and so are even doing the vertical
       | integration themselves. In such an event does genAI end up being
       | the automation tool that you access/use through OpenAI or google
       | and that's it?
       | 
       | I am sure people here see it better than I do, so what new class
       | of problems is this genAI going to solve?
        
         | dmos62 wrote:
         | It's not too expensive. I did some crawling the other day. AI
         | wrote the crawler automatically, then I had it extract
         | structured info from the crawled product info. It was about 600
         | requests to o4-mini. Cost me about 2.5 cents.
        
           | occamschainsaw wrote:
           | o4-mini? It is a rumored model building up on o3, and
           | different from 4o-mini. Sorry, their naming scheme is a mess.
           | Would be cool if you do have o4-mini access though.
           | 
           | https://www.theverge.com/news/646458/openai-gpt-4-1-ai-model
        
             | dmos62 wrote:
             | My mistake, I meant GPT-4o mini.
        
         | fhd2 wrote:
         | Well, what's a tool? I would say:
         | 
         | 1. It solves a problem. Doesn't have to be a completely
         | unsolved problem, can just be a new solution. Or even just new
         | packaging on an old solution. But it needs to solve some kind
         | of problem.
         | 
         | 2. It's trustworthy. Some people get a tool to suite their own
         | process. But the majority, from anecdotal evidence, will adopt
         | the tool's process. There's this idea that "these guys know how
         | to do invoicing so I don't have to think about invoicing if I
         | use their invoicing tool".
         | 
         | 3. It's known. A bit philosophical, but if something exists
         | that nobody _knows_ solves a problem they might not even know
         | they have, how much of a useful tool is it, really?
         | 
         | DropBox is an interesting example. It wasn't exactly a major
         | scientific breakthrough, and a lot of people asked "why don't
         | people just use FTP?". If you focus on (1), DropBox looked
         | close to pointless. But what they did is nail (2) and (3).
         | 
         | Now, if you subscribe to the hype, you might argue (1) and (2)
         | will soon be covered. AI will magically solve your problem and
         | be a universal domain expert telling you what to do, so you
         | don't have to think about it. You might also argue that it will
         | magically solve (3), with stuff like Gemini Live kinda watching
         | you all day and constantly going "let me tell you how to do
         | that" or "let me do that for you".
         | 
         | Seems unlikely to me. Not impossible, most things I can think
         | of are theoretically possible. Just unlikely. And if you think
         | even just _one_ of those three aspects can't be fully automated
         | in the near future, there's still plenty of opportunity left to
         | differentiate in that area.
         | 
         | I think generative AI does unlock a new generation of startups,
         | because it's genuinely new technologies that we can find at
         | least some valuable use cases for. And an army of startups
         | tends to be better at quickly exploring a new solution space
         | than a few big incumbents. So in that sense, it is similar to
         | smartphones, which also brought a new solution space, and with
         | it, startups.
        
           | bryanrasmussen wrote:
           | the problem for me with AI startup, built on someone else's
           | AI, is that it is difficult for me to envision what the moat
           | will be - perhaps because it is difficult to think up what
           | the moat will be for something so generically specified, but
           | really whenever I think of something cool, it seems AI is an
           | added on ability but not the main driver, and that AI does
           | not give any moat whatsoever.
        
             | fhd2 wrote:
             | I agree, it's an implementation detail. What you sell is a
             | solution, and your mode is generally your brand, your
             | existing user base, and any anti competitive shenanigans
             | you manage to get away with. Not so much even the actual
             | quality of the product, it's more important what people
             | _think_ it's quality is.
        
           | inferiorhuman wrote:
           | It's trustworthy.
           | 
           | In what way is AI trustworthy? It's ruining the parts of the
           | internet I use and care about. I can't visit Digikey's site
           | in an incognito tab without having to sit through a ~5 second
           | captcha these days. Mouser is less aggressive, but it's still
           | problematic. Drew's spent how much time combating AI bots
           | instead of improving Sourcehut?
           | 
           | In fact I'd be hard pressed to think of a site that isn't
           | getting boned by AI.
        
             | fhd2 wrote:
             | Well, I meant it in the sense of: You use a tool because
             | you think it's creators really know this space and you can
             | rely on it. To stick to the DropBox example, their users
             | probably trust them to keep file management convenient for
             | them and to keep their files safe.
             | 
             | I've seen a depressing amount of people treat LLMs like
             | some sort of oracle. So I can picture a significant number
             | of consumers just trusting ChatGPT with their taxes or
             | whatever, based on the assumption that it know any domain
             | at least as good as human experts.
             | 
             | I'm not saying _I_ find any LLM trustworthy. But if enough
             | people do, it becomes difficult to differentiate there.
        
         | motoxpro wrote:
         | I have no idea if this will be as big as an iPhone, but the
         | early iPhone apps and development followed a similar
         | trajectory. Viral flashlight apps and others that then got
         | integrated into the OS, similar to API wrappers and viral photo
         | makers of today getting swallowed up by the next model update.
         | 
         | In terms of what problems it solves, I would imagine that will
         | be up to the developers/companies to come up with the
         | Uber/Airbnb/Tiktok, that the iPhone enabled, that AI enables.
         | Same as any platform.
        
         | n_ary wrote:
         | Here is a different take. The GenAI is just another revenue
         | stream for big players with scale or resources.
         | 
         | At work, I was tasked with building some magical agentic stuff,
         | working on it for a while, I realized that, HN shouts,
         | oAI/xAI/Google/Amazon/Anthropic have no moats, there are oss
         | models available. The actual fact is, the moat is the access to
         | scale of resources(GPUs, power infra, network), which is very
         | difficult to build for random joe's startup.
         | 
         | You must always rent a model access from one of these
         | players(even OpenRouter delegates), and that is the exact moat.
         | 
         | GenAI solves problems of having more generalized solutions, so
         | instead of a super customized secret sauce solutions for your
         | product as competitive edge, you now build magic prompts for
         | GenAI to take the same input and hopefully with enough GenAI
         | ingesting the same data and coming to consensus, you get a
         | reasonably useful output that previously your custom solution
         | was capable of. Since you no longer have a custom solution, you
         | now pay off the GenAI operators(the real moat of GenAI for all
         | operators hosting it). In the process,you also sacrifice your
         | competitive edge of that super secret IP and relying heavily on
         | GenAI prompting correctly and numerous verification in each
         | step with enough automation, which of course again costs money.
         | 
         | GenAI is the new hammer of visionary leadership and executives
         | (a hefty amount of money has been burned to campaign and PR to
         | convince these people) to use it everywhere, so the operators
         | can ensure that they make some profit from the amount of money
         | they sank on it. If you super impose the "AI" of current year
         | to "Apps" in 201x era, where everything must have an "app",
         | you'll suddenly realize that we've seen the same before and of
         | course most apps need cloud... and as clouds have costs, apps
         | became more subscription model instead of previous 200x era buy
         | once, use forever.
        
           | samvher wrote:
           | If the only moat really will be the scale of computation
           | resources, that's great news for users, because it will be an
           | extremely competitive market where prices will be driven down
           | very effectively.
           | 
           | I suspect that model quality/vibes and integrations will play
           | a role as well though.
        
         | TeMPOraL wrote:
         | I personally _don 't want_ to see "a whole new generation of
         | startups". GenAI, LLMs in particular, are a powerful tool for
         | the users precisely _because_ they are general.
         | 
         | After all, what is the business of such startups of the "AI
         | age"? It's using the AI models on the backend, where users
         | can't reach, to sprinkle some magic onto features. But as a
         | user, I don't want your product to use AI as a tool, I want
         | _your product to be a tool the AI uses_. The former keeps power
         | away from users, the latter gives it back to them.
        
         | monkeydust wrote:
         | I do some angel investing so get a lot of deal flow in my
         | inbox. A lot of the AI ones (which is now 99% of everything)
         | are not really businesses.
         | 
         | They are at best nice features or capabilities to have in wider
         | enterprise application suite for say HR or Accountancy but on
         | their own its just a lot of smart people working extremely hard
         | to sell one tiny solution that is small cogwheel for a much
         | larger problem.
         | 
         | My gut is telling me that very soon, if not now, there will be
         | an opportunity for savvy VCs to sweep up some of these
         | cogwheels and package them together into a real business and
         | it's something I am exploring with a few other angels. Curious
         | what others think on this. Feel free to DM me (details on
         | profile).
        
           | spacebanana7 wrote:
           | Won't the general purpose models eventually eat all the
           | application layer business models?
           | 
           | Now that ChatGPT desktop can read files in my code editor and
           | apply changes I've pretty much stopped using dev specific AI
           | tools. Same with spreadsheet problems - where uploading data
           | to ChatGPT and hoping for the best actually works pretty well
           | now.
        
           | fxtentacle wrote:
           | I fully agree with this.
           | 
           | I was involved very early with automated speech recognition
           | for transcribing meetings, but then both UberConference and
           | Google Meet just integrated it into their existing offerings,
           | thereby massively reducing the market size for standalone
           | solutions. And given how heavily subsidized AI API calls are
           | at the moment, just relying on them is a huge risk for your
           | business model, because you never know when your suppliers'
           | prices will 10x to represent the true cost of providing those
           | AI services.
           | 
           | In my opinion, the sales of many of these new AI tools are
           | mostly driven by the existing audience of the creator. In
           | many cases, you could just ask an LLM to quickly build you
           | your own competing solution, which you can then use for free.
           | E.g. all those SEO content and LinkedIn blog post bots. Vibe-
           | coded AI "businesses" are the software equivalent of branded
           | white t-shirts.
        
         | ijidak wrote:
         | It's too early to tell. We're only 24-36 months in.
         | 
         | We're still in the post Netscape, pre-dotcom-crash bubble.
         | 
         | Real applications are coming.
        
         | BeetleB wrote:
         | > However genAI is too generic and too blunt a tool, as in it
         | does everything.
         | 
         | How does this preclude the AI "age"? And why is the metric
         | "companies make money off of it"?
         | 
         | I view it more like open source/Linux. When Linux was new, it
         | was immensely useful , but not a means for companies to make
         | money (except for a tiny few).
         | 
         | Or more precisely, far more people used Linux for their own
         | personal benefit than companies making money off of it.
         | 
         | And so it is with generative AI. For me, personally, it's
         | _very_ useful.[1] So assuming the major companies (OpenAI,
         | Anthropic, etc) don 't go bankrupt and/or kill it off, gen AI
         | is here to stay, and will continue to disrupt. That startups
         | can't make much money off of it is somewhat irrelevant.
         | 
         | [1] It has revolutionized speech recognition for me. I have
         | several automations running that were coded by Claude. Things
         | that had been in my mind for years but I didn't have time to
         | write them. MCP will add more value to my life soon. Etc.
        
       | Timwi wrote:
       | I don't understand why Android only? Why would nobody want to use
       | this on a PC?
        
       | pcdoodle wrote:
       | No thanks google.
        
       | kleiba wrote:
       | Cool tech, but for some reason, the very first sentence in every
       | reply the AI gives in the demo videos is really off-putting to me
       | personally. It seems to me like perhaps this kind of joyful
       | helpfulness introductory sentence is expected in US culture, but
       | it immediately turned me off. I come from a culture that is less
       | verbose in general and more to the point, so this feels like a
       | mismatch right away.
        
       | terminatornet wrote:
       | the first example video tells you how to improve your home decor
       | by saying that you could add a side table and blanket to your
       | chair. Thank you Gemini for telling me that tables can go next to
       | chairs.
       | 
       | There's really no use for AI outside of making studio ghibli
       | drawings and giving me a bunch of broken code quickly.
        
         | netsharc wrote:
         | The answer seems to also be very surface-level, wow it sees a
         | chair and lamp in a corner and believes it's the perfect
         | position, without apparently considering anything else in the
         | room.
         | 
         | Marie Kondo doesn't need to worry about being replaced by AI,
         | yet...
        
       | 1024core wrote:
       | Is there a way to record the screen WITH AUDIO and save it?
        
       ___________________________________________________________________
       (page generated 2025-04-11 23:02 UTC)