Hacker Newsnew | past | comments | ask | show | jobs | submit | mitchell_h's commentslogin

I take it as a signal they don't see any of their value being provided by the models. They're strong point was never frontier technologies. It's always been the delivery of the technology.

I use anthropic's models daily, and sometimes switch to Gemini. Google is losing the marketing front BADLY, but their AI service is surprisingly great. It's far cheaper than anthropic for one. and for my kind of research it's just better.

I'm quite certain that Google's AI services are likely the most used in the world right now by virtue of having the widest distribution. It's in the search box. It's on your Android phone. Just because they aren't the preferred coding or research agent does not mean they are losing - that's a pretty small slice.

It can be everywhere, but that doesn't mean users are paying or even value it.

See also: Windows / Notepad / M365 / GitHub / Paint / Xbox / Azure / Solitaire / D365 / Security Copilot.

Yeah this seems true. Claude Code are famously dubbed as best AI coding agent, but google doesn't care about that niche I guess. Somehow, I still rely on google search as they have diversified it.

If you ask questions, it will enable "AI overview" , but if we search about particular object/platform like "Google stock" or "bbc news", it will give the old classic search experience and we woulnd't need to swallow "AI overview" pill in that case.


I tried using Gemini CLI to sort some code issues for me, ran out of tokens mid-way through, even though I have Gemini Pro.

Turns out licensing is separate for "code" and "pro"...


Same happened to me. That was the death knell for Gemini as a coding agent to me. I even paid for a whole year...

I highly suspect they opaquely lowered usage limits on me.


Is it? My mom and all her friends use "the intelligence". What is it? Gemini, because it's on their android phone.

Apple played a blinder by calling it "Apple Intelligence".

Well done lads


who cares about marketing when you have distribution? Probably a smart move to pump dollars into the product and not the marketing.

in high margin businesses, customer acquisition is everything.

If your product becomes commoditized, it’s no longer a high margin business

> If your product becomes commoditized

Depends on the product - whether protein bars, salty chips, cellular service, or IPhone or something else. If your product has a flavor, it’s never going to get commoditized. Coke still tastes better than Pepsi.


This is the power of a brand. Kirkland and some private label products are literally the same as the competitor products and yet are perceived differently. Even in your Pepsi vs Coke example, Pepsi routinely wins in blind taste tests but there are more "Coke people".

It will be interesting to see if the LLM companies can establish their own "brand" and how they will do that. LLM voice is a thing but not sure if it's a good thing people will use to hang their self identity on. Distillation of models and constant training also make this complicated. Claude code is winning on harness and ux right now but it seems precarious and also easy to commoditize. I think elon tried to add branding to his chatbot pretty intelligently by being iconically crude/evil/"anti woke" since it's both highly visible and less likely to be copied.

We live in fascinating times!


You can have a high accounting margin and a product with price equal to economic marginal cost—externalities, cost of capital, barriers to entry… DRAM is a commodity but has (currently) a high margin.

And Google acquired you in 1998 with search.

flash 3.5 is the best price/performance model for what i'm doing. I had been using opus for everything but as we started running many agents at once, and then eventually agent managing sub agents frontier is not an option.

we started model testing the cost/performance of our skills and agents and flash 3.5 wins in most things.

As people develop harnesses for their codebase i think the intelligence required comes down a lot.


I have not tried the Gemini CLI in a few months but when I did it was a shit show.

Google makes it very hard to use their shit and it was full of bugs.

Anthropic's current run is based entirely around Claude Code in this space and the last time I used the gemeini-cli it wouldnt give me access to the latest models and I was paying them for the privilege


Google trashed the Gemini CLI client and replaced it with agy (antigravity), which is written in go and is much nicer.

Interesting you say that. Every user I speak to says antigravity cli is missing lots of features and Gemini cli was working quite well. Same for me.

It's not as feature rich, but has also not crashed once for me, unlike gemini cli, which was a flickery, unstable mess.

So they did.

https://github.com/google-gemini/gemini-cli/discussions/2727...

I get the complaints in that thread but I still think it is hilarious. That repo is a gong show to random shit and perhaps one of the best worst examples of "opensource" LLM development.


It will also just sit there "thinking" for ages, if whatever you are doing requires an input (like sudo)

Sometimes you have to tab across and give it a PW, but it seemingly is incapable of parsing that, and just asking.

Kiro, what we use at work, on the other hand will just prompt you. (And doesn't like taking credentials directly)


We use Kiro (AWS) and Gemini (Google) at work.

Kiro is of course really good to back into AWS stuff, it knows more about AWS than Amazon themselves!

Gemini is really good at understanding my inane ramble and mis-spelling


I think Google is a bit sandbagging here knowing they have all the data and likely better models hiding. My theory is it's a bit of not disrupting the stock market direction by exposing whose really the boss. If they can do it cheaper, faster, and better, people start asking questions, especially with upcoming IPO's.

This makes no sense. Google is beholden to its own shareholders, not the markets at large.

In any case, it's well known that devs in Google have liked anthropic/openai models for coding more than gemini, so unless they're hiding their best models from the people within, I think it's just the case that they're behind.


It's more that they know they can eventually clone any successes the other companies have and steal their market share. Their really is no moat. In a more normal environment they would be buyout candidates but that's a bit too far gone at this point, so you just let them run until they are out of gas and Google can benefit from any advances without upfronting the cost.

Even with anthropics record breaking revenue growth I don't see how the pure AI companies can sustain, but the catch-22 is that any obvious pivot proves that. This puts the more traditional tech companies in position to ride the back of the wave until the growth curve tops.


> they know they can eventually clone any successes the other companies have

Google has gone all in on AI. To the point of challenging their own core product. Apple is waiting and seeing. Google is building and distributing, albeit with terrible marketing.


Apple isn’t waiting and seeing on the hardware side, only implementing AI on the software side, which there doesn’t seem to be much of a demand for them to do. Apple are well set for on-device LLMs and agents with their Mx Max cpu/gpu, and their wait on the rest is saving them hundreds of billions by not burning all their profitability to the ground building Nvidia-filled datacenters the same as everyone else, which is why Google is now having to hunt for extra money by raising capital like this.

search is not their core product though, it's ads. they ain't challenging anything.

Ads are meaningless without a surface to show them on. Search is absolutely a core product.

Coding is a pretty small slice of the markets in play. Google's models are driving cars right now. Using coding agents doesn't give much insight into performance in the broader world; I would assume assume Google is performing better in general even if Claude or Codex is currently outperforming for coding.

> Coding is a pretty small slice of the markets in play.

I don't think that's true, mostly in that a lot of usecases are solved via coding models + a harness.

> Google's models are driving cars right now.

Yes + other models like alphafold. But those are (relatively) specialized models. Besides, the comment I was responding to was saying Google is sandbagging the market to keep it calm or something. I don't disagree that Google is doing well overall and has some clear advantages


Google also owns 15% of anthropic.

Pedantic correction that doesn't change anything other than accuracy: it was reported over a year ago to be closer to 14% than 15%.

https://www.theverge.com/news/627849/auto-draft

But I believe since then Anthropic have raised more money, almost certainly diluting Google's stake (I could be wrong and misremembering that Google didn't partake in the additional fundraising). I have in the back of my head that Google is down to something like 10% now, but don't have time to go and find details to fact check that, sorry!


It's important to remember that the cloud division, rapidly becoming Google's golden goose, does not give one fuck about Gemini and would happily sell out all of Gemini's compute to Anthropic and OAI if given the opportunity.

I think it's proper. When you release something like this, a raw data dump is the only way to cut out a BUNCH of the "this is modified and falsified" noise.


Yes. Importantly just because they've processed it conveniently doesn't mean they'd ever intend to share that.

My first thought when I saw this is how much will it cost me to kick it up to a HF I stance.

I did a trial run with the Epstein files and it was genuinely fun to catch a few bits before the media caught up.

Not to mention that if they add any metadata thats just increasing their exposure and they will be held to what the LLMs label it.


The login cards were the killer feature on them(at the time). I managed a fleet of them things spread all over 4 buildings. Being able to work in one location, get up and goto another and just pickup what you were doing was INSANE in that day and age. Slapping in a keycard do it all was unheard of.

We had citrix and sunray in those days. Citrix was for those that had BIG BIG BIG money and needed windows. We were a java shop, so it was either an e450 in the server room and sunrays, or ultra5s at every desk.


I never got to use this, but it seems like it would hit my dream computing environment (which has since advanced to an idea that my phone would fill that role so I could be working on my phone, plug my phone into a desktop or laptop workstation, continue working there, unplug, continue on the phone, move to another computer, etc. Apple’s Handoff almost scratches that itch, but it’s not quite as reliable or ubiquitous as I would like and the ideal would be that I have my whole working environment portable via the phone.


I never used it, but Microsoft's Continuum [1], was supposed to scratch that itch too. Your phone could drive a desktop experience when you connected (wired or wireless) to a docking station (I saw a laptop shaped dock which might have been a prototype), and with the proper implementation of UWP apps (which didn't really happen, afaik) you could interact with your apps/data equally in desktop and mobile. Didn't let you run win32 apps though, which makes it kind of limiting, but if all you do is browser, messenger(s), and office suite, it could have worked pretty well. I think this would have worked better with Intel's x86 phone cpus, but those were cancelled days before the Continuum reveal, and Microsoft also did a really poor job on WM10, so nobody knows about any of this.

[1] https://en.wikipedia.org/wiki/Windows_Continuum


The real question is whether the phone is the actual compute, or if it's just the key card that lets you access the remote compute. I think that (provided sufficient connectivity) both of those can be reasonable trade-off options, but they certainly present different views of the world.


Oh man, SunRays and e450s, desktop sessions that ran 24/7 on beefy servers accessible from anywhere, U5s with those type-V membrane keyboards... Every detail of your post makes me warm and fuzzy with nostalgia. :-)

(Except the Citrix. I never admin'ed that, only used it for a few gigs.)


I saw those in use 20 years ago, when physics-class visited the local nuclear-science research center. It felt like Sci-Fi then and I have not yet seen this replicated anywhere else sadly...


Nice series. The real problems created by microservices need a solution.


They can claim that...but if you've built a public SaaS before you know the job is not to host the software, it's to put rails around people taking it down. They've had since 2008 to build those rails, and they're just now hitting places that take the service down on the regular?


The problem is that they are charging per seat and need to start charging for usage.


> I imagine the clutch is easier on the knees these days! Modern tractors don't really have a clutch. I mean they sorta do, but it's electronic. Even on sizable consumer positioned tractors(I have a JD 5055, but it applies to almost all the JD models), there's just a lever for forward, N, and reverse. Gear shifters work MUCH MUCH better now.


When I was younger I absolutely HATED changing gear on the tractor - it was a matter of dropping the revs which caused a dive, then a clunk finding the gear, then a jolt as the gear took hold and the revs came back up

I never felt in control of that old beast


Changing gears while driving? Are you sure you where supposed to? Many old tractors are without synced drives, so you are supposed to select gear before you start driving. Of course you can change when driving, but then you have to match revs to not get the drop betwen


You have to match revs anyway. Without Synchromesh the trick is to double declutch.


Ha ha, that's such a wonderful description, that's exactly how it feels!


I watched some explain how deepseak got good and the Chinese approach to LLM training. Really wish I could remember it. The premise was China thinks of LLMs not as a thing separate from hardware, but gains efficiencies at each layer of the stack. From Chips to software, it's all integrated and purpose built for training.

Wonder if Anthropic is making a mistake by focusing on "consumer" hardware, and not going super specialized.


So you watched some random video from some random YouTuber, didn't even remember who made it, so much so you didn't even remember that deepseek isn't spelled "deapseak", didn't bother to even find it or verify, and then you go asserting your memory as fact on a serious discussion forum.

Comments like yours add nothing to the discussion.


I belive he does have a valid point.

You can throw money and hardware at a problem, but then someone may come along with a great idea and leapfrog you.

Just consider that all major AI providers now use deepseeks ideas for efficient training from that first paper.


thank you for the aerious discussion my good sir I tip my hat to you


DeepSeek uses merchant silicon like everyone else.

edit: I misunderstood, I thought you were implying they designed their own GPUs. nevermind


> I watched some explain how deepseak got good and the Chinese approach to LLM training.

I distinctly remember reading a big pantie twisting from Sam Altman and Co that Chinese took their stuff, the stuff OpenAI and Co spent billions to create, and used that as the base for $0.00


It’s fake news predicated on China not being able to get GPUs. But it turns out everyone was getting them their GPUs by serial number swaps in warehouse.


In my AGENTS.md I have two lines in almost every single one: - Under no condition should you use emoji's. - Before adding a new function, method or class. Scan the project code base, and attached frame works to verify that something else can not be modified to fit the needs.


I'm curious about the token usage when it scans across multiple repositories to finding similar methods. As our code grows so fast, is it sustainable ?


I'm sold. My wife is a teacher(refuses to use a chromebook because they suck, or an Ipad because data input sucks). She'll be getting one of these. Kids probably will too. I'm having a really hard time finding another laptop on the market the hits the "i use gmail, and gdocs, and some other webapps all day" demographic so well.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: