The keynote was the most boring for me. I paused it to go to the bathroom, and even forgot that I was watching it.
I think they've lost track of the meaning of IO and its keynotes for users. They should rather have a separate Gemini event like they had a separate Android one last week.
They're collectively losing track of their product verticals because they're too focused on shoving AI down everything.
Google Home is a cluster-f, basic things keep failing, and every other announcement from the Google Home VP is about Gemini. It took them years to reduce the frequency at which devices go offline.
Even their sessions seem underwhelming. It's a mixture of "what's new in X" and "AI" this and that.
Watching these bland presentations with choreographed delivery and reading off a prompt off-screen (I'm not completely sure they're doing this, but it looks like it) makes me appreciate Steve Jobs presentations from the past so much more.
Steve really had product presentations down. I wish people at least tried to copy him.
This is AI-maxing bleeding into everything. “Make me a great conference presentation”, and this is roughly what you get. AI-nshittification writ large. It’s sad to watch.
I wonder what's in store for the local Gemma models, as well as Flutter. I've been making fully local apps that either download Gemma 4 2B or use the built-in AICore in Android and Apple's Foundation Models. Local models are getting really good these days including web search and tool calling such that for many use cases I don't even need cloud models.
It must have improved considerably since I tried the "3.5-flash-preview" a couple of months ago if all these claims in the presentations are true. Because it couldn't even make changes in a 200 line Python script without doing major mistakes (like messing up argument order when calling functions) when I tried it.
It's super hard to know if those prices are reflective of the true cost.
Remember that leaderboard position is very important, and many leaderboards are perf/$. So, to push the share price up and be top of leaderboards, the company might falsely quote a loss-leading price, and maybe set quotas so people can't cause too big losses.
flash beating the pro it was distilled from is suspicious, not surprising.distillation usually loses you something. if the smaller model is winning on agentic evals, the more likely read is the evals weren't measuring agent quality in the first place. that's the bigger problem for builders, not which model to pick.
Unironically you can put "what are good demos for agentic workflows at Google I/O that would be received well by the general public" into Gemini's AI Mode and get better suggestions for use cases than what they're showing.
In the US, wired phones were often leased from the phone company! If you shop for vintage phones from the 1960s, they will often have a decal to that effect. Examples:
I wasn't alive in the 90s, so for me it was 2012-2015. Android was pretty open, iOS was easier to jailbreak, and the ecosystem felt a lot "freer" than today where you step outside of Google or Apple's garden and get endless Cloudflare captchas, apps refuse to load due to attestation, and things are designed without privacy in mind.
They claim the AI itself built an OS and demoed Doom running on it. Personally, rather than Doom, I think I'll only truly appreciate the real value of this advancement when flawless real-time YouTube subtitle translation actually becomes a reality.
> They claim the AI itself built an OS and demoed Doom running on it.
Doom ran on MS-DOS, which - by modern standards - provided a shockingly minimal set of abstractions for programs. I think about the only thing you need to run Doom is the int 21h "API" to access the FAT filesystem and perform keyboard I/O. Note that MS-DOS did not provide facilities such as memory virtualization / management, process management, video drivers, sound drivers, etc - that was all provided by the hardware itself, which had its own hardware interrupts handled by the code in the device's ROM. It's why Doom required you to choose the type of a sound card you have, the interrupt / DMA channel to use, etc.
So I think this is a lot less of a flex than it seems; in fact, Anthropic using agents to build a semi-unusable compiler is far more of an achievement. Providing enough of int 21h to run Doom is probably something that a human could do in a weekend, doubly so if they can peek at the source code of FreeDOS.
You have a real knack for explaining things. That makes total sense. Regarding the compiler part, I'm actually trying to build my own language using LLMs, and to be honest, it barely works. I wasn't very familiar with classic OS architectures like MS-DOS, but your breakdown really helped me understand the context
One can only dream of Gemini Spam-Detector 3.5, which will finally make YouTube useable again. I can’t believe the kind of crap that gets recommended in the Kids app.
I don't think you can get much more real-time than what we already have? iOS and Android both do live translations through their respective earbuds at decent latency, you can't get much faster without running into fundamental linguistics limitations.
For instance, you can't translate a Japanese sentence into English until you reach the verb at the end; no amount of latency improvement can overcome the fact that languages have different word orders.
You're right, but the YouTube subtitle translation actually isn't working right now. The English captions show up fine, so reading them isn't an issue, but stil
> On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals.
Ugh. Was really hoping that they were going to actually announce a release date for the display glasses this year at I/O, but it seems that they still don't have the tech ready for this. Audio only this year is a massive disappointment.
How would unoptimized slop code be good for older hardware? That's what Linux and other projects are for. If you wanted to tinker and learn how operating systems work, you'd code one yourself like I did in CS classes. You'd actually learn something and you get the good feeling of having done it yourself.
You won't understand a thing, though. You already have the source code for Linux, Minix and tons of other OSes, pedagogical or otherwise. Why generate yet another that you won't understand?
Because it would be sponsored by anthropic/google/openai? You cannot do it (typically) without paying for the tokens they only can offer. Programming used to be free, but slowly, we need to pay for every single line of code. It's sickening
Only speaking for myself, but I use it a lot, and intentionally. Enough that I set up a search engine shortcut for it in my browser (g <space> type prompt here <enter>).
I much prefer it to having to click through links to find things. My last handful of searches were:
- Looking up open hours for a local store
- Defining words
- "postgres select where string has prefix"
- "cloudformation read parameter from ssm"
Things where I want to look up a fact, but want an answer right away without having to read through multiple pages.
I agree, the AI overview is definitely worse. I'm talking specifically about the AI mode search (at https://www.google.com/search?udm=50&aep=11). The AI overview seems to be summarizing the search results that were returned for your query already, while AI mode seems like it's doing its own searches based on your query.
I would definitely give it a shot if you haven't tried it before.
I just gave it a shot and it hallucinated an off-trail, alpine scrambling route that, based on my knowledge of the area, would potentially get someone killed. I simply asked it if it could "talk to me about X route", and then it decided to make shit up instead of just regurgitating the information in the sources it cited.
It hallucinated how long a scree slope was, made up the existence of a "high rocky knoll", and insisted someone could traverse via a heather slope that's actually non-existent.
I've used it for live service video games, it's pretty good at summarizing major changes to a game since you've played it last. With regular web search you'd have to go to every major patch and a lot of games don't even have good patch notes / it's all stuck in content creator videos.
Though I still prefer Claude for this since it's better at citing sources.
Not surprising. It’s placed exactly where the regular search results used to be (when navigating away from image results) and muscle memory is strong. Haven’t clicked it intentionally once though.
Yeah I actually use AI mode a fair bit. It has access to Google's index so it can be quite a good search engine interface when the normal one doesn't work (which is quite often).
Yeah I mean Big Tech are using a lot of it because they're training models and shoving AI into everything. But if they weren't forcing it upon people, would that same demand be there?
It's discouraging to see Google price Gemini 3.5 Flash at 3x the cost of Gemini 3 Flash. I would think that most people that deployed this model in production would have used it for low latency tasks, classification/categorization, customer support or basic RAG-/RAG-style chatbots. Performance on coding benchmarks is nice and all, but where is the "intelligence too cheap to measure"? This new cost point is quite prohibitive and will eat up a lot of margins if developers adopt it.
Expect all models to increase in price 3x with new releases. They're easing us into the margins they're targeting.
Flash 3 wasnt appropriately priced, it was priced to get you used to a certain level of spending, then they'll crank it up and get you used to the next level of spending.
I am aware that it was likely subsidized or at least did not have appropriate margins. But over time, that same capability should become profitable if parameter efficiency and chips improve. For many customer facing use cases outside of coding assistants, optimizing for speed, basic logic/maths and conversational texts matters much more than being able to use 40 tools simultaneously. I would have hoped that Google would recognize this and keep a dual line up, where Pro and Flash are clearly intended for different market segments. But it seems, it's all in on coding assistants and screw the other use cases..
Now, we might need to change to DeepSeek 4 Flash if Google deprecates 3 Flash.
When using Hetzner, DigitalOcean or any other VPS service together with Cloudflare, I can handle millions of page views for 5-50$ a month at pricing that has stayed nominally the same for a long time and due to inflation and performance gains of the underlying chips has basically become cheaper.
I guess you didn't get the memo from last month: Loss leader pricing is over, you're now paying a less subsidized price, and will continue to until it's profitable
As explained in another comment, I think this is more about Google orienting Flash towards more complex use cases. If we got minor improvements vs 3 Flash with 1.5x the price so they can optimize their margin (which on such small models for conversational tasks is a completely different stories than the 3-25x subsidies that these agentic coding plans offer) I would have been happy. Or even no change at all. But knowing Google, I now must fear that they will deprecate 3 Flash without offering any realistic option for that user facing chatbot segment that does not require multi-tool use across 500k context.
I was curious and just installed it, and... Antigravity is a literal clone of VSCode. Wtf? Honestly, it's so embarassing. I might write a blog post about this, but I remember falling in love with the art of product watching Google demo Google Wave. Janky sure, ahead-of-its-time maybe, but also visionary and mind-blowing. Here we are almost two decades later and Google is re-releasing something made by Microsoft. The epitome of laziness and uninspired hive-think.
Imo, there's so much room for an actual normie end-product that supercharges local work with AI for regular people (office workers, creatives, etc.), but a VSCode clone ain't it. (Insert: fine, I'll do it myself Thanos meme.)
You have to realize why something like antigravity exists.
There are two usual ways it occurs:
1. Political fights internal to the company resulting in incoherent strategy and products. HN assumes this is almost always the case, and but it's only sometimes the case :).
or
2. A bunch of execs sitting in a room saying stuff like "we have to have a platform with eyeballs that we control where we can surface our AI innovations and tools or else we'll be disintermediated/unable to release stuff that matters" or whatever.
(or both!)
The second part is often a real problem to solve. The first (you have to have a platform) does not follow.
At least two of the main issues with solving these kinds of problems this way (ie antigravity) is:
a. No user actually cares about your strategic problems and isn't interested in helping you. What you release still has to be valuable/etc enough that people are willing to use it over their existing tooling. At least right now, antigravity really isn't.
b. The strategy seems to assume a complete vacuum where it's Google vs existing tools. However, there are tons of large developer companies with the same exact problem of wanting a place they control to surface stuff (or whatever particular problem this is meant to solve). If they opt for the same approach, why would Google's strategy beat them? . If they opt for a different approach, same question.
If you poke there, i suspect you will find nobody has good answers to these questions.
So this approach turns into, at best, skating to where the puck is instead of where it will be.
Nowhere on their marketing copy do they own up to that. Even the majority of the UI screenshots intentionally exclude the full UI look and feel and most are plucked out to not even look contextually like they're part of a greater IDE interface. It very much feels like they want to call it their own and not a fork of VS Code.
They have like 100 junk AI products, antigravity is just one. AI is hardly super charging anyone's work either, regardless of how you package it, especially normies.
95% of people just want to search things and be entertained with their tech. Most dont want to write slop emails at light speed or whatever it is you think AI might be useful for to the average person.
If you consider slopifying your output at a really high velocity "supercharging" then maybe.
Kinda annoying how Google always releases new products "in a safe and secure way" to a handful of "trusted testers". They already fumbled their image generation launch with Imagen a couple years ago while DALL-E rolled out in the ChatGPT app, and likewise with video generation. Took a while to regain the mindshare with nano banana. With the new Spark and stuff locked behind "trusted testers", I'm worried that again they will get overtaken by competitors while waiting in the name of "safety".
I wondered why they updated the Gemini Chat modes today, removing "Thinking" and adding "Thinking level" to Flash. Looks like marketing has been working overtime.
can't wait to hear all about the new Agentic workflows you can build with AI and build with agentic agent swarms and build more workflows to do more with Enterprise AI with Gemini Enterprise Agent Platform and elevate efficiency with the power of AI and unlock value with AI agents. also can't wait to hear more about AI integrations across Google AI Workspace, like AI Gmail or AI docs (powered by Gemini Enterprise AI). also can't wait to empower myself with more AI-powered data that will be unlocked with agents. AI.
I can't think of a way to mock all the ai hype more than what they actually just talk about.
"Next, our hacker news AI agent reads and comments for you based on your commenting history, no need to think or do anything at all - we'll automate your time wasting and make better, more relevant jokes than you & your karma score will increase exponentially."
Are you working for hire? I'd like you to join my team where we are automating parametrizing paradigms to foster growth in changing paradigms of parametrizing workflows.
yes I am. I prefer a modern 996 schedule to maximize efficiency. let me know how I can contact you and we can set up a meeting to discuss collaboration and elevate our synergies to grow revenue through AI powered velocity
I went to every single I/O since the very first one, but stopped going last year after finding it utterly uninteresting due to this fixation. It's great hearing about new tech, but it felt like every single session was either about AI or had AI awkwardly crammed into it - even sessions where it had nothing to do with the core subject matter. It was like some gatekeeper told the engineers their topic wouldn't make the lineup unless the word "AI" was in the title.
Honestly getting tired of this knee jerk anti-AI stance here on HN. We’re talking about the most revolutionary technology in history that’s going to change every facet of our lives. It’s getting toxic, now we have people getting booed at graduations for just mentioning AI. There will be a transition but it’s not the end of the world. Don’t cling to the past, look forward to the future.
While I get your point, this kind of gives me "Old man yelling at cloud" vibes. Yes, all the AI talk and bullshit bingo became quite annoying at this point, and I also can't wait for it to settle. But AI is here, and it's here to stay. Wether we like it or not. It's like what dotcom was for the internet back then. We'll get through this eventually - with a bubble bursting here and there - but making fun of it with overtuned phrases like "Everything will be connected to the internet in the future, even your fridge, car and toothbrush!" won't age too well I am afraid.
I think the point is that AI was here 40 years ago [1].
LLMs/RAGs/Transformers are the newish thing that's here to stay.
I've seen my colleagues vocabulary regress from "training transformers" to just "using AI", without clarifying if are using claude or actually building a network. I was recently told that no one says "vibe coding" any more (now it "agentic AI", I was told). My colleague who does ML research was told he was the only one at his workplace that wasn't doing AI.
So the problem isn't the technology (a lot of the technology is great), it's that the discussion around it has been dumbed down by hype.
and I'm just regurgitating what Google sends me via email and funny things like renaming Vertex AI to "Gemini Enterprise Agent Platform" (not a joke) even though Vertex is mostly used for inference, e.g. Claude via gcloud or fine tuning models etc.
and I use Claude code every day, so I'm not like completely dismissing AI/agentic stuff.
Chewy.com is not pets.com. They're in the same market, but it's not useful to say that a business built around limitless hypergrowth is the same as a business built around medium sized sustainability.
It seems pretty clear to me that we've exhausted the possibilities of the transformer architecture. Whatever we're using in 20 years will certainly be a different technology.
The point is that other topics exist that deserve talking about. There is SO much talk about LLMs everywhere, and in this kind of event they will eclipse other, perhaps more interesting conversations.
I could see this image in my mind before I clicked on it, before even consciously considering what it might be. Is this how an LLM feels on the inside?
Antigravity seemed to work well at first, but the same model on the same software now seems to fail to edit most files most of the time, and then get itself tied in knots trying to resolve the error by editing files with awk, sed, grep, etc!
Yup. “Wait, let me circle back and fix this another way.”
What they can never fix is that plenty of Pro users complain that they never get quota, models are always maxed out. I left, and I can’t believe how much time I wasted in AGY steering Gemini or reminding it that no, you can’t install random new dependencies or disable tests.
I cant see how they could. The Gemini cli repo was a shit show the last time I looked a month ago and the service itself wouldnt even let me use version > 2.5 even though I was a paying customer.
It's two slightly different jokes right? The parent comment joke is essentially stating "they are definitely going to talk about AI a lot" and the second joke is stating "they are not going to talk about anything besides AI", which are similar but technically different.
Don't give them the idea, before too long they'll retire the email part so you could use it to train agents and use AI in your workflows by leveraging Artificial Google Intelligence (AGI)
ASI now. You'd think contrasted with AGI it would be Artificial Specialized Intelligence, but it's Artificial Super Intelligence. Since AGI already happened and didn't mean anything, the next step is a super AI.
I still want the normal definition of AGI... Which my understand is: no more re-training, real time memory and re-learning. The fact they cannot fit a lot of context, or step into "we no longer need to train it" territory tells me, they're farther than they keep pretending they're close to.
I actually love this. They think they'll be able to control this tech and be lords over everyone. In the meantime everyone is replacing them with homegrown solutions.
Two years ago everybody was explaining that Google was done, that because of AI search was dead and that they were the IBM or the 2020s for they were absolutely nowhere went it came to AI.
Now we're at a point where a little flash model from Google is SOTA on half of the benchmarks:
So the tune changed: Google is now dead not because they'd be nowhere in AI but because they're too good (?) at AI?
So basically: whatever happens, this time for Google it's over right?
(not too clear why it's a 122 Kb .gif file as if the 90s called over dial-up modems while when the same in .webp would be less than half the size but I digress)
If you don't think AI will include sponsors/ads/etc once someone comes out on top, well, I might have a bridge to sell you.
Seriously though, I'm not sure why Google evolving in this manner precludes them from having a profitable business model. Right now we're subsidizing the costs (probably just a bit) and having ongoing subscription revenue they can increase as needed (particularly in the "google won the race" scenario) will be key before they even have to consider layering advertising on top.
I think the difference this time is, anybody can curate a baseline set of training data, there is no need to constantly be scanning the open web and indexing it. Everyone already has "good enough" question-answering capability. There is no option to pay or an ad-free, trackerless search funciton on google, but I can do that with multiple non-google providers. Between LLM and kagi I've managed to largely cut google out of my life for $40/mo. Subsidized LLM will eventually disappear, but I think cost per token will reduce over time to meet the ~$20/mo ad-free tier.
advertising is ~76% of Alphabet revenue. Cloud is 12% and growing 30%+, but margins arent comparable yet — search basically prints money, cloud is still scaling to prove it.
But google search has subpar quality for many queries compared to ChatGPT and other AI providers. Even if they did fix the quality issue, nobody has yet got a good way to integrate paid ads within an LLM response.
Funny how you are saying this now, but not 2 to 3 years [0] ago when OpenAI and Microsoft was eating Google's lunch and even 9 months ago when everyone on this site thought Intel was "dying". [1]
Now you want to buy Intel and Google just as I am about to sell it you at 9-10x and at a 3x multiple?
I think they've lost track of the meaning of IO and its keynotes for users. They should rather have a separate Gemini event like they had a separate Android one last week.
They're collectively losing track of their product verticals because they're too focused on shoving AI down everything. Google Home is a cluster-f, basic things keep failing, and every other announcement from the Google Home VP is about Gemini. It took them years to reduce the frequency at which devices go offline.
Even their sessions seem underwhelming. It's a mixture of "what's new in X" and "AI" this and that.
Steve really had product presentations down. I wish people at least tried to copy him.
Fart noise
Are they just training it a bit longer until it tops benchmarks?
Remember that leaderboard position is very important, and many leaderboards are perf/$. So, to push the share price up and be top of leaderboards, the company might falsely quote a loss-leading price, and maybe set quotas so people can't cause too big losses.
Is it? I thought Flash 3.5 was beating 3.1 Pro.
https://i.ebayimg.com/images/g/71gAAeSw1URp8QPF/s-l1600.webp
https://i.ebayimg.com/images/g/tPAAAeSw~RxqBiLR/s-l1600.webp
I think this started to change only in the 1980s and 1990s, thanks to cheaper electronics imported from overseas?
Doom ran on MS-DOS, which - by modern standards - provided a shockingly minimal set of abstractions for programs. I think about the only thing you need to run Doom is the int 21h "API" to access the FAT filesystem and perform keyboard I/O. Note that MS-DOS did not provide facilities such as memory virtualization / management, process management, video drivers, sound drivers, etc - that was all provided by the hardware itself, which had its own hardware interrupts handled by the code in the device's ROM. It's why Doom required you to choose the type of a sound card you have, the interrupt / DMA channel to use, etc.
So I think this is a lot less of a flex than it seems; in fact, Anthropic using agents to build a semi-unusable compiler is far more of an achievement. Providing enough of int 21h to run Doom is probably something that a human could do in a weekend, doubly so if they can peek at the source code of FreeDOS.
For instance, you can't translate a Japanese sentence into English until you reach the verb at the end; no amount of latency improvement can overcome the fact that languages have different word orders.
The YouTube real-time translation are utter garbage. And that may the less bad of the real-time translations out there. Still pure garbage.
If you watch anything specific to one sport/hobby, the number of words that are incorrectly translated is just wild.
1. A new Gemini model that is not quantized.
2. A way to connect NotebookLM notebooks to any agents if you pay for the pro subscription.
> On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals.
https://developers.googleblog.com/an-important-update-transi...
Could be a good use for older hardware. Why not.
I much prefer it to having to click through links to find things. My last handful of searches were:
- Looking up open hours for a local store
- Defining words
- "postgres select where string has prefix"
- "cloudformation read parameter from ssm"
Things where I want to look up a fact, but want an answer right away without having to read through multiple pages.
I would definitely give it a shot if you haven't tried it before.
It hallucinated how long a scree slope was, made up the existence of a "high rocky knoll", and insisted someone could traverse via a heather slope that's actually non-existent.
Though I still prefer Claude for this since it's better at citing sources.
Flash 3 wasnt appropriately priced, it was priced to get you used to a certain level of spending, then they'll crank it up and get you used to the next level of spending.
Now, we might need to change to DeepSeek 4 Flash if Google deprecates 3 Flash.
Why would you expect text-generator-as-a-service to be any different?
https://antigravity.google/pricing
this works: https://www.youtube.com/watch?v=wYSncx9zLIU
Imo, there's so much room for an actual normie end-product that supercharges local work with AI for regular people (office workers, creatives, etc.), but a VSCode clone ain't it. (Insert: fine, I'll do it myself Thanos meme.)
There are two usual ways it occurs:
1. Political fights internal to the company resulting in incoherent strategy and products. HN assumes this is almost always the case, and but it's only sometimes the case :).
or
2. A bunch of execs sitting in a room saying stuff like "we have to have a platform with eyeballs that we control where we can surface our AI innovations and tools or else we'll be disintermediated/unable to release stuff that matters" or whatever.
(or both!)
The second part is often a real problem to solve. The first (you have to have a platform) does not follow.
At least two of the main issues with solving these kinds of problems this way (ie antigravity) is:
a. No user actually cares about your strategic problems and isn't interested in helping you. What you release still has to be valuable/etc enough that people are willing to use it over their existing tooling. At least right now, antigravity really isn't.
b. The strategy seems to assume a complete vacuum where it's Google vs existing tools. However, there are tons of large developer companies with the same exact problem of wanting a place they control to surface stuff (or whatever particular problem this is meant to solve). If they opt for the same approach, why would Google's strategy beat them? . If they opt for a different approach, same question.
If you poke there, i suspect you will find nobody has good answers to these questions.
So this approach turns into, at best, skating to where the puck is instead of where it will be.
Nowhere on their marketing copy do they own up to that. Even the majority of the UI screenshots intentionally exclude the full UI look and feel and most are plucked out to not even look contextually like they're part of a greater IDE interface. It very much feels like they want to call it their own and not a fork of VS Code.
But wait, there's more ... you can also view https://antigravity.google/product/antigravity-2 where it's no longer a VS Code fork but now a clone of Claude Desktop!
I think it's a fair to label this all as unoriginal and uninspiring.
95% of people just want to search things and be entertained with their tech. Most dont want to write slop emails at light speed or whatever it is you think AI might be useful for to the average person.
If you consider slopifying your output at a really high velocity "supercharging" then maybe.
https://www.voxelbench.ai/leaderboard
Better leaderboard: https://arena.ai/leaderboard/text
And the order of the top players is very different...
All 3 major labs streamlined their desktop apps and plans to be the exact same. And we are still doing email drafts as "consumer use cases".
As I'm typing this they are talking about reimagining the search box. They turned it into a chat window.
Truly the pinnacle of innovation.
Google is trying to kill off the open web here. They tried before, e. g. AMP and what not.
I think it is now time for all of us to help retire Google. This planet can only take some Evil - and Google just exceeded planetary limitations.
"Next, our hacker news AI agent reads and comments for you based on your commenting history, no need to think or do anything at all - we'll automate your time wasting and make better, more relevant jokes than you & your karma score will increase exponentially."
Fire. Electricity. The wheel. The printing press. Assembly lines. Flight. The computer. Space exploration.
Anyone else care to contribute?
LLMs/RAGs/Transformers are the newish thing that's here to stay.
I've seen my colleagues vocabulary regress from "training transformers" to just "using AI", without clarifying if are using claude or actually building a network. I was recently told that no one says "vibe coding" any more (now it "agentic AI", I was told). My colleague who does ML research was told he was the only one at his workplace that wasn't doing AI.
So the problem isn't the technology (a lot of the technology is great), it's that the discussion around it has been dumbed down by hype.
[1]: https://archive.org/details/byte-magazine-1985-04-rescan
and I'm just regurgitating what Google sends me via email and funny things like renaming Vertex AI to "Gemini Enterprise Agent Platform" (not a joke) even though Vertex is mostly used for inference, e.g. Claude via gcloud or fine tuning models etc.
and I use Claude code every day, so I'm not like completely dismissing AI/agentic stuff.
Remember HN is mocking the capability/technology itself not the ability of specific firms to survive.
Seems like a stretch, considering it's down 50% compared to a year ago.
https://www.nyse.com/quote/XNYS:CHWY
I just.. I don't know the mental model of the people who speak like this. What is the point you are trying to make..
Your writing says a lot about you, too.
What they can never fix is that plenty of Pro users complain that they never get quota, models are always maxed out. I left, and I can’t believe how much time I wasted in AGY steering Gemini or reminding it that no, you can’t install random new dependencies or disable tests.
It's not the favorite, but it's definitely "working."
> It's constantly overloaded and shits the bed
Or happen avoid all the downtime.
Claude has one 9 of uptime, downtime for me multiple times per week. [1] When it works, it's great. It just doesn't work that often recently.
[1] https://status.claude.com/
$INTC and $GOOG are good buys right now!
Two years ago everybody was explaining that Google was done, that because of AI search was dead and that they were the IBM or the 2020s for they were absolutely nowhere went it came to AI.
Now we're at a point where a little flash model from Google is SOTA on half of the benchmarks:
https://storage.googleapis.com/gweb-uniblog-publish-prod/ori...
So the tune changed: Google is now dead not because they'd be nowhere in AI but because they're too good (?) at AI?
So basically: whatever happens, this time for Google it's over right?
(not too clear why it's a 122 Kb .gif file as if the 90s called over dial-up modems while when the same in .webp would be less than half the size but I digress)
Seriously though, I'm not sure why Google evolving in this manner precludes them from having a profitable business model. Right now we're subsidizing the costs (probably just a bit) and having ongoing subscription revenue they can increase as needed (particularly in the "google won the race" scenario) will be key before they even have to consider layering advertising on top.
But google search has subpar quality for many queries compared to ChatGPT and other AI providers. Even if they did fix the quality issue, nobody has yet got a good way to integrate paid ads within an LLM response.
I'd say those are 2 huge risks to their business.
Now you want to buy Intel and Google just as I am about to sell it you at 9-10x and at a 3x multiple?
[0] https://news.ycombinator.com/item?id=39527133
[1] https://news.ycombinator.com/item?id=48066995
Canceled my $20/mo tier.
Two prompts took me into 67% usage. One of those prompts was lost completely and errors out when try to access it.
Gemini users are livid.