I recently discovered just how much food engineering goes into products like potato chips and similar items, not only in the flavor, but also in the texture and even the sound it makes.
There's someone at the factory who basically spends their time eating samples from new batches to ensure they're exactly as expected.
Same! Somehow I had never heard of this. Will have to try and visit someday!
Also, lol @ the name of the river near that site. (For the gringos: "Rego Grande", "rego" officially means ditch/trench, but nowadays it's much more commonly used as slang for butt crack. So, it's the Large Butt Crack river. Which I guess tracks for Brazil :P)
Just because Anthropic and OpenAI really want there to be an arms race justifying the outsized investment, doesn't mean the optimal play is to build larger, more expensive, models.
The capital infusion the frontier labs have received has gotten to a size where many believe it may not be possible to recoup this investment without some very unrealistic things happening.
I think it's reasonable to not completely drain one's cash reserves trying to stay ahead in a race where participants may very clearly be about to run straight off of a cliff.
And it doesn't have to be either/or. They could make larger, more expensive models, just at a slower cadence.
Sure downside would be not learning from people using your model for coding, if we're on the cusp of huge leaps in self-improvement. But there is a reasonable case for avoiding desperate scramble, especially if other parts of the business can also create value with the compute.
If the Chinese labs can compete on a shoestring budget with access to much less powerful hardware, Google should be able to compete as well. They're becoming almost irrelevant for agentic coding right now.
I would agree with you on AI/LLM being more than agentic coding but at the same time, I think there's more nuance.
For example, PDF's and powerpoints can be generated using agentic coding by things like https://bento.page or other ways of generating them in an agentic coding fashion.
A lot of browser automation could/is also done by agentic coding.
It can also help them set up and configure self hosted software with the help of LLM's and debugging if its working or not.
You can create videos using Manim and remotion.dev and also excalidraw-animate and generate excalidraw files agentically if what you need is more vector style graphics (which surprisingly can fit into many ideas) rather than say a real life human waving video/more photo-realistic video (but I must say that this has certainly its own pros/use-cases as well).
It might sound self-explainatory but turns out that coding can represent a wide range of problems!
I get that. I use agents a lot and LLMs often reason with code. It is valuable. I just think the floor is a lot lower for general reasoning and common tasks like that. And in 6-12 months it won’t matter. Google will publish better models. The temporal distortion of how long a Sol or a Fable has existed is real. No one is suddenly missing out on some giant competitive edge because their model is a few months behind. I feel like it’s all just going to normalize and things other than how well your model can write code will matter more and more in 12 to 24 months.
Sure I understand what you mean as well and I am not asking for SoTA models to be created by Google but more so explaining why coding is still the largest focus for many labs.
I personally wish to get more smaller models (like the recent qwen model) and other open source models like GLM 5.3 and the glm flash model.
> No one is suddenly missing out on some giant competitive edge because their model is a few months behind
Sure I can agree with that. The competitive edge might still exist but I do get the underlying sense of what you are trying to suggest.
> things other than how well your model can write code will matter more and more in 12 to 24 months.
What are the things then which you feel like could be more differentiative factor? For example, I personally think multi modal is still quite preferrable in AI models. I use GLM 5.2 and it doesn't have vision and I can certainly imagine time/use-cases where multi-modality would've helped coding and even other use cases as well. So what are some other use cases that you are thinking? Video generation models like Veo/Sora?
It's not much of a shoestring budget to be receiving regular injections of investment from state lenders along with cheap credit.
I don't think the comparison holds.
Yes, I agree with you that the race all the AI companies are running doesn't make sense, but at the same time, there are rumors that Google has produced newer versions of Pro without releasing them to the public.
Version 3.1 has plenty of room for improvement, yet they don't seem to be giving the attention it deserves or at least communicating accordingly.
There is more to the cost of a model than its training.
While training is a significant Capex expenditure, it has very low Operational cost after training unless it is deployed for public inference.
It may be that they wish to slow their cadence of releases, or develop their models to focus more in a different direction, etc. No matter what the actual reasoning, they have chosen to not compete in the same race, and I cannot say I fault them.
I work there. I have zero internal knowledge about the model. Opinion my own, etc. I don't think it is worth fighting to win on a month to month time horizon. When you step back and look an inch above this market, Gemini Pro 3.1 as a product was released in February. 6 months. It feels like forever and that Google is behind, but on a 2-3 year horizon? The models are going to stay similar.
Also, look at Flash 3.5 to 3.7. Flash 3.7 is a genuinely decent Sonnet 5 class model. Flash 3.7 is quite efficient too. Also, whatever was spent training 3.5 pro is probably not wasted. However, as a strategy, when I see models like Kimi K3, Fable, Sol. If you discard "because the model sucked" what other alternatives or potential options might exist?
I thought of a quite a few and they are far more compelling and interesting to me.
(Also Gemini models tend to be pretty decent at more than just programming. Enterprise AI use is more than just software eng / programming)
I'm the CTO of a GCP shop with an 8 figure annual commit.
If you'd told me at the end of Cloud Next 2025 that by now Google still wouldn't have a competitive offering to agentic coding offerings from Anthropic (Claude Code + Fable) or OpenAI (Codex + Sol), I wouldn't have believed you.
In our non-coding use cases where we're embedding models in our product, we're also not reaching for GCP stuff. Because Anthropic has the mindshare of our engineers and product folks, since it's what they use every day.
Given your position and the responsibility that comes with it; I sure hope you updated your mental model in another way than simply "they are acting irrational"... It's not clear from your comment that you did, but it sounded a bit like it.
3.5 was almost certainly a 3.1 post-train, so likely a small investment on Google's part.
They mentioned that they have already started pretraining Gemini 4, which will be the full ground up rip-your-face-off-expensive training that is often discussed.
Google doesn't have a good coding model. This is a HUGE problem. They don't need "larger more expensive models", they need a good coding model because it's a competitive advantage.
Second tier models are like self-driving cars to the point where you question if they save any time. Sol can do a deep analysis and plan a large feature. Luna can mostly execute. There's a clear qualitative difference.
I have a pro subscription, I think they have just given up. Likely because when they test their new models against the other frontier models they are so bad, they just pull it back. This leads them to try and innovate in other areas where there is currently less competition so they can compete. Not a bad play.
In the real world out there, Google and Microsoft are absolutely dominating enterprise customers.
Every single non-tech office worker I know is writing Gemini "gems" (sort of claude prompts/skills) or prompting Copilot to help drafting board meeting notes, insurance contracts updates that reflect changes in regulations, make quick loan feasibility assessments before passing them to the relevant office, presentations, etc, etc.
I'm talking insurance, banking, consultancy, manufacturing, etc, etc.
Why? Because Google and Microsoft already were in these companies, all they had to do is "oh, you also have AI now with your plans". Procurement and data compliance are the first thing businesses have to sort out. They were already sorted out.
Google doesn't need to have the best coding model or triumph in meaningless benchmarks, it only needs their models to get better and cheaper while serving them to their existing customer base.
They are playing a different game.
And Microsoft, doesn't even need to care about models at all, they can provide whatever open or closed AI with their services and have to focus on the harness in Excel or Github/Azure Copilot or whatever.
E.g. while developers in most of my clients use whatever they prefer or the company pays for, the remaining 90% uses either Google or Microsoft products.
Not a single one has incentives into venturing into OpenAI or Anthropic or Z.Ai lands because they might be better at some benchmark that is completely irrelevant to their tasks of updating powerpoints or summarizing incoming emails.
Google is an advertising company with an enterprise SaaS branch. I bet they have chosen to focus on running the most efficient "everybody" model instead of running a heavy model for coders.
When using Gemini for other tasks than coding, it is actually pretty good. It grounds well with Google search and gives mostly correct, well written answers to many niche questions.
I think they also have the problem of having given away their pro subscription to 10s or 100s of millions of students worldwide. They're tightening down on that now, and I have a feeling that this goes into them not releasing a larger model.
They blew up my interest when the stole my money by cutting me off from Gemini CLI with no explanation or recourse. I did not violate the terms of service and my only crime seemed to be not wanting to use Antigravity. They still took my money for the rest of that month and gave me nothing for it.
> It will probably destroy this battery very quickly.
'Very quickly' seems to be measured in decades from what I see
Most people I know have their laptop docked, either near-permanently at home or taking it between home and a desk with dock (so it's nearly always plugged in when in use)
The 'servers' I've had were all old laptops - yes, with battery (don't have to buy a separate UPS if one is already included). I've now also had two non-server laptops that were plugged into power pretty much all the time. I treated my first laptop perfectly: taking out the battery anytime I'd be in a classroom for several hours and it was already full enough (you know, back when sliding out your battery was possible in the first place), but I'm not noticing any faster degradation on the systems I treat poorly. I'll activate the 80% cap function where it exists (that is, in the latest phones and my work laptop's UEFI config panel), but the evidence of it being helpful has yet to materialize
I have read that studies show that charging to 100% causes degradation and I trust those more than random people's anecdotes (and I'm a random person), but I'm starting to wonder if I've maybe just misunderstood the study results. Maybe it's like how smoking increases your chance for some cancers but you might also be fine in the end. I see people with plenty of much younger and better-treated batteries that can't sustain literally 30 seconds of use whereas mine are reasonably close to the new state (some haven't had many charge cycles, just sat at a stressy high voltage all the time, which was indeed supposed to destroy them), and I don't know how to explain that otherwise
I had a laptop as a server for a really long time, and I also poked around using a phone for some server things (not as main server) and after some time it was unusable (the whole phone was suffering because wasn't prepared for a battery that fucked).
Anyway, my point is, maybe his phone has some mechanism similar to laptops to prevent this kind of thing, but in the end it's a phone. It's not the intended use for it.
Well, maybe, but could that not also be the chance thing I mentioned? That it greatly increases odds, but a 'great increase' of 1% per decade is still not more than a third of the cases or so. Or it's the bypass charging that a sibling comment mentioned indeed
Some phones do too, it depends on their power management integrated circuit (PMIC) and driver. Though it's always better to physically disconnect the battery when not in use, as there is always a trickle current otherwise (most "spicy pillows" batteries are due to deep discharge, as far as I know). The best practice is to store batteries disconnected, at 60% capacity). In theory you could build a small circuit to sit in between the phone and battery to protect it, maybe I'll have a go at it.
What you can do, with some extra steps, is have Home Assistant installed (and have a Home Assistant server, I guess!) on the phone and use a smart power switch that's also associated with Home Assistant.
The phone's battery can be reported to Home Assistant, where an automation can be triggered to turn off power to the phone when the battery gets above a certain percentage and turned back on when it goes below.
Not the best solution, but certainly an alternative from messing around with trying to get the phone to work without a battery.
I came here to complain about something similar, the excessive use of box shadows on both the top and the sides. It made the text horrible to read on mobile, if it not for Firefox's Readability mode.
I saw the weird subtle pattern and first freaked out that my monitor's backlight was failing (high flicker rate revealing LCD refresh patterns). I also was playing a video with the browser in the background and it was choppy, and I first thought it was my video player messing up until I realized the web page had some pointless background animation.
I just tested again and even my cursor gets jump over the page. Amazing they'd do this for such a subtle, useless effect.
Switched from firefox few months ago, I don't like google, but firefox has only few percent of market share currenly, many pages simply do not work properly with it, plus it has bugs on macos (like onmousover stuff), which simply make it unusable. Safari is fine, but also many websites (which sometimes you need to use like banking or gov) are not tested on it. The overall browsers situation is less than ideal.
That is weird, I've daily driven firefox for the better part of a decade (aside from when my employers have required chrome) and seldom encounter issues. I'm curious where you're hitting these. In fact, since ublock origin got removed from Chrome, the experience is far better on firefox.
Also, if everyone chooses to not use firefox because it has low market share, it'll remain low market share forever.
> many pages simply do not work properly with it, plus it has bugs on macos (like onmousover stuff),
I use firefox on mac and I have simply no clue what you're talking about. Tons of people use firefox on mac quite successfully...
The only page that I know of that doesn't work is google earth - it doesn't work on linux either. (Technically it does, just incredibly laggy compared to chrome)
I can't think of a single macos specific bug. Or a single mouse over related bug.
I think it's a stretch to call firefox on mac "unusable". Like once a month I'll have something not work, and half the time it also doesn't work in chromium (both are the developer's fault). And when it does work in chromium, browsers are free so just switch over there and switch back. And to top it off, there are endless free flavors of chromium like brave etc, that virtually never have compatibility issues without needing to go full google.
I have been using Firefox for years (decades) and I rarely encounter a page or site that doesn't work... a handful perhaps, but it's a complete non-issue for me. If anything, my ad-blocker can sometimes cause issues and I'll disable that and everything is good again.
Technically speaking, Safari is the second most used browser. It’s quite surprising how often I get these “unsupported browser” popups. I still use it as my primary and have FF as a backup. No Chrome needed.
I've been using Firefox pretty much exclusively since its first release (I did a year or two on Opera about a decade ago), and I can't remember it failing to render any site at all. I do run across broken sites from time to time, but they always work once I dial back my extreme privacy settings.
Really? I use Firefox as my personal browser and everything works fine, including Google sites. Very rarely there’s a government site that needs Chrome, but I definitely wouldn’t say it’s “many” sites.
There's someone at the factory who basically spends their time eating samples from new batches to ensure they're exactly as expected.