There's all sorts of things you can do if you don't care about money.
The more interesting point is that if you aren't driven by investors to care about short term financial stuff (stock prices) then you can make long term decisions. Caring about your customers is a classic one for this - costs you money in the short term, but in the long term gets you a great customer base.
They care about money. They definitely care about money. They have achieved a steady cash flow that can sustain their business forever, unless something really bad happens.
What they don't care is the endless growth that MBA guys always try to achieve, and the quarterly profit driven decision making that ultimately destroys their customers loyalty, for short term profit.
A business can be very profitable without being exploitative. It's the people in Wall Street who can't seem to understand this. For them a hundred million dollars of profit is good if last year it was only fifty million dollars, and a dying business if last year it was also a hundred million dollars. It really makes no sense.
Just thinking out loud, but I wonder if Wall Street would be less awful about ruining companies if we were able to get a more meaningful dividend out of your average company? So perhaps the stock price itself stays relatively flat or boring, but the dividend paid out makes up for it. Or perhaps it would be the exact same issue and they’d be squeezing companies to maximize dividends.
I just know that I expect stock prices to go up because most “dividend stocks” give such a small amount of money per share.
This is the magic of the decentralized, invisible-handed, "free" market. Nobody (in particular) tells you what to do, and (ideally) you reach a canonical equilibrium which may (under some idealised circumstances) be optimal (in some sense).
If ifs and buts were candy and nuts this would be the cat's pyjamas. I shan't deny it's mathematically elegant, and also feels good in many ways, but the real trouble is it's exceptionally hard to form a watertight argument for an alternative.
Put another way, the appeal of the free market isn't so much in its correctness as it is in its simplicity. I can personally attest that it's sumple enough for any fool to understand, in an area of economics where it's devilishly difficult to establish anything solidly.
I say all this as someone who is a big fan of Valve and their work, deapite otherwise being a foss zealot, just because they throw a bone to our sort.
My impression of this is that it is partially a tax policy issue.
Dividends are taxed differently and higher than capital gains. So given a choice between a stock buyback and a dividend, often a buyback makes more sense.
People who are capable of saying "I have enough now" will self-select out of the activist investor class. Automatically, the people with the most power to influence publicly traded companies will be people who demand endless growth.
This sort of thing is why I think we need heavy taxes to limit wealth accumulation. Money is power, and the amount of power a person with ten+ figure wealth wields is too much for any one person to have, let alone one who was never elected.
This is sortof a function of buybacks vs dividends. Like, if the market rewards growth (in terms of share prices) then line must go up forever. If you are getting a steady stream of inflation adjusted cash (i.e. dividends), then you can afford to care less about number go up.
Why not do a CI pipeline from the beginning instead of relying on trust that no one ever forgets to run a check, considering adding CI is trivial with gitlab or github.
Because it adds friction, and whoever introduces that CI pipeline will be the one getting messages from annoyed developers, saying "your pipeline isn't working again". It's definitely a source of complexity on its own, so something you want to consider first.
U agree it adds a bit of complexity, but all code adds complexity.
Maybe interacted with CIs too much and it's Stockholm syndrome, but they are there to help tame and offload complexity, not just complexity for complexity'a sake
> they are there to help tame and offload complexity, not just complexity for complexity'a sake
Theoretically. Practically, you're hunting for the reason why your GitHub token doesn't allow you to install a private package from another repository in your org during the build, then you learn you need a classic personal access token tied to an individual user account to interact with GitHub's own package registry, you decide that that sounds brittle and after some pondering, you figure that you can just create a GitHub app that you install in your org and write a small action that uses the GitHub API to create an on-demand token with the correct scopes, and you just need to bundle that so you can use it in your pipeline, but that requires a node_modules folder in your repository, and…
Oh! Could it be that you just added complexity for complexity's sake?
Uh, I've used the package repository with GitHub, and I don't remember having to do this! So, I'm not entirely sure what's happening here. I think this might be accidental complexity because there's probably a misconfiguration somewhere...
But on that point I agree, initial set-up can be extremely dauntin due to the amoun of different technologies that interact, and requires a level of familiarity that most people don't want to have with these tools. which is understandable; they're a means to an end and Devs don't really enjoys playing with them (DevOps do tho!). I've had to wear many hats in my career, and was the unofficial dedicated DevOps guy in a few teams, so for better or worse had to grow familiar with them.
Often (not always) there's an easier way out, but spotting it through the bushes of documentation and overgrown configuration can be annoying.
I'm aware of how much overhead CI pipelines can be, especially for multiple platforms and architectures, but at the same time developing for N>1 developers without some sort of CI feels like developing without version control: it's like coming to work without your trousers on.
Yeah, that was my entire point really—there's some complexity that's just warranted. It's similar to a proper risk assessment analysis: The goal isn't to avoid all possible risks, but accepting some risk factors as long as you can justify them properly.
As long as you're pragmatic and honest with what you need from your CI setup, it's okay that it makes your system more complex—you're getting something in return after all.
Because then you're wasting time trying to quote bash inside of yaml juuuust right to get the runners to DTRT.
Okay no but seriously, if you're not being held back by how slow GitHub CI/Gitlab runners are,
great! For others they're slow as molasses and others in different languages with different build systems can run an iteration of their build REPL before git has even finished pushing, nevermind waiting for a runner.
If you used the first gen "pay later" services they'd scrape you for "compliance checking" or simply mask it as a transaction which is actually just personal information scraping.
Most of the times you did not see it, as it's obfuscated as a part of the transaction.
They are also the companies complaining a lot about the "failure" of the PSD standards since it limits how much and how obfuscated they can scrape everything (and there are records).
This is such a funny workaround, I like that. But it doesn't matter in any library or your own code, since factory methods are much better (simply because they have names).
From what I've seen, most people only care about the LTS versions of Java. Which means that after Java 21 LTS comes Java 25 LTS. The same happens with Ubuntu (after 22.04 LTS comes 24.04 LTS).
Why wouldn't you factor in training? It is not like you can train once and then have the model run for years. You need to constantly improve to keep up with the competition. The lifespan of a model is just a few months at this point.
In a recent episode of Hard Fork podcast, the hosts discussed an on-the-record conversation they had with Sam Altman from OpenAI. They asked him about profitability and he claimed that they are losing money mostly because of the cost of training. But as the model advances, they will train less and less. Once you take training out of the equation he claimed they were profitable based on the cost of serving the trained foundation models to users at current prices.
Now, when he said that, his CFO corrected him and said they aren't profitable, but said "it's close".
Take that with a grain of salt, but thats a conversation from one of the big AI companies that is only a few weeks old. I suspect that it is pretty accurate that pricing is currently reasonable if you ignore training. But training is very expensive and the reason most AI companies are losing money right now.
Unfortunately for those companies, their APIs are a commodity, and are very fungible. So they'll need to keep training or be replaced with whichever competitor will. This is an exercise in attrition.
I wonder if we’re reaching a point of diminishing returns with training, at least, just by scaling the data set. I mean, there’s a finite amount of information (that can be obtained reasonably) to be trained on. I think we’re already at a sizable chunk of that, not to mention the cost of naively scaling up. My guess is that the ultimate winner will be the one that figures out how to improve without massive training costs, through better algorithms, or maybe even just better hardware (i.e. neuristors). I mean, we know that at worst case, we should be able to build something with human level intelligence that takes about 20 watts to run, and is about the size of a human head, and you only need to ingest a small slice of all available information to do that. And training should only use about 3.5 MWh, total, and can be done with the same hardware that runs the model.
> But as the model advances, they will train less and less.
They sure have a lot of training to do between now and whenever that happens. Rolling back from 5 to whatever was before it is their own admission of this fact.
I think that actually proves the opposite. People wanted an old model, not a new one, indicating that for that user base they could have just... not trained a new model.
That is for a very specific class of usecases. If they would turn up the sycophancy on the new model, those people would not call for the old onee.
The reasoning here is off. It is like saying new game development is nearly over as some people keep playing old games.
My feeling: we've yet barely scrarched the surface on the milage we can get out of even today's frontier models, but we are just at the beginning of a huge runway for improved models and architectures. Watch this space.
which is completely "normal" at this point, """right"""? if you have billions of VC money chasing returns there's no time to sit around, it's all in, the hype train doesn't wait for bootstrapping profitability. and of course with these gargantuan valuations and mandatory YoY growth numbers, there is no way they are not fucking with the unit economy numbers too. (biases are hard to beat, especially if there's not much conscious effort to do so.)
Does the cost of good come down 10x or not? For say Uber it didn’t, so we went from great $6 VC funded product to mediocre $24 ride product we have today. I’m not sure I’m going to use Copilot at $1 per request. Or even $0.25. Starts to approach overseas consultant in price and ability.
well, Uber always faced the obvious problem of scaling (even after level 42 self-driving, because it's not possible to serve local demand with global supply, plus all the regulatory compliance issues - which they initially "conveniently" sidestepped by being bold/criminal, but cities are not going to play dumb forever)
of course these chat-AIs also started by "well maybe it's fair use", but at least the scaling problem seems easier than for taxi services
I suspect we've already reached the point with models at the GPT5 tier where the average person will no longer recognize improvements and this model can be slightly improved at slow intervals and indeed run for years. Meanwhile research grade models will still need to be trained at massive cost to improve performance on relatively short time scales.
Whenever someone has complained to me about issues they are having with ChatGPT on a particular question or type of question, the first thing I do is ask them what model they are using. So far, no one has ever known offhand what model they were using, nor were not aware there are more models!
If you understand there are multiple models from multiple providers, some of those models are better at certain things than others, and how you can get those models to complete your tasks, you are in the top 1% (probably less) of LLM users.
This would be helpful if there was some kind of first principle at which to gauge that better or worse comparison but there isn't outside of people's value judgements like what you're offering.
I may not qualify as an "average user" but I shudder imagining being stuck using a 1+ yr stale model for development given my experiences using a newer framework than what was available during training.
Passing in docs usually helps, but I've had some incredibly aggravating experiences where a model just absolutely cannot accept their "mental mode" is incorrect and that they need to forget the tens of thousands of lines of out of date example code they've ingested during training. IMO it's an under-discussed aspect of the current effectiveness of LLM development thanks to the training arms race.
I recently had to fight Gemini to accept that a library (a Google developed AI library for JS, somewhat ironically) had just released a major version update with a lot of API changes that invalidated 99% of the docs and example code online. And boy was there a lot of old code floating around thanks to the vast amounts of SEO blog spam for anything AI adjacent.
>Passing in docs usually helps, but I've had some incredibly aggravating experiences where a model just absolutely cannot accept their "mental mode" is incorrect and that they need to forget the tens of thousands of lines of out of date example code they've ingested during training. IMO it's an under-discussed aspect of the current effectiveness of LLM development thanks to the training arms race.
I think you overestimate the amount of code turnover in 6-12 months...
Strangely, I feel GPT-5 as the opposite of an improvement over the previous models, and consider just using Claude for actual work. Also the voice mode went from really useful to useless “Absolutely, I will keep it brief and give it to you directly. …some wrong annswer… And there you have it! As simple as that!”
The "Pro" variant of GTP-5 is probably the best model around and most people are not even aware that it exists. One reason is that as models get more capable, they also get a lot more expensive to run so this "Pro" is only available at the $200/month pro plan.
At the same time, more capable models are also a lot more expensive to train.
The key point is that the relationship between all these magnitudes is not linear, so the economics of the whole thing start to look wobbly.
Soon we will probably arrive at a point where these huge training runs must stop, because the performance improvement does not match the huge cost increase, and because the resulting model would be so expensive to run that the market for it would be too small.
>Soon we will probably arrive at a point where these huge training runs must stop, because the performance improvement does not match the huge cost increase, and because the resulting model would be so expensive to run that the market for it would be too small.
I think we're a lot more likely to get to the limit of power and compute available for training a bigger model before we get to the point where improvement stops.
As long as models continue on their current rapid improvement trajectory, retraining from scratch will be necessary to keep up with the competition. As you said, that's such a huge amount of continual CapEx that it's somewhat meaningless to consider AI companies' financial viability strictly in terms of inference costs, especially because more capable models will likely be much more expensive to train.
But at some point, model improvement will saturate (perhaps it already has). At that point, model architecture could be frozen, and the only purpose of additional training would be to bake new knowledge into existing models. It's unclear if this would require retraining the model from scratch, or simply fine-tuning existing pre-trained weights on a new training corpus. If the former, AI companies are dead in the water, barring a breakthrough in dramatically reducing training costs. If the latter, assuming the cost of fine-tuning is a fraction of the cost of training from scratch, the low cost of inference does indeed make a bullish case for these companies.
> If the latter, assuming the cost of fine-tuning is a fraction of the cost of training from scratch, the low cost of inference does indeed make a bullish case for these companies.
On the other hand, this may also turn into cost effective methods such as model distillation and spot training of large companies (similarly to Deepseek). This would erode the comparative advantage of Anthropic and OpenAI, and result in a pure value-add play for integration with data sources and features such as SSO.
It isn't clear to me that a slowing of retraining will result in advantages to incumbents if model quality cannot be readily distinguished by end-users.
In the same way that every other startup tries to sweep R&D costs under the rug and say “yeah but the marginal unit economics have 50% gross margins, we’ll be a great business soon”.
Google and Facebook had negative free cash flow for years early in their lives. All the good investors were lolling at the bad investors lolling at the cash they were burning.
Ok and lets compare the cost of running those products and reinvestment vs the model businesses.
FCFF = EBIT(1-t)-Reinvestment. The operating expenses of the model business are much higher - so lower EBIT.
The larger the reinvestment the larger the hole. And the longer it continues (without clear steep barriers to entry to exclude competitors in the long run) it becomes harder to justify a high valuation.
I really dislike comparisons like this - it glosses over a lot of details.
One can explain the equation all they like - the fact is that negative free cash flow is just a reality of the early stages of some very, very good businesses.
In the 90's and early 2000s, but people laughed at businesses like Amazon & Google for years. These types of people highly focused on the free cash flow of a business in it's early years are just dumb. Sometimes a business takes a lot of investment in the early stages - whether it's capex for data centers or S&M for enterprise software businesses, or R&D for pharma businesses or whatever.
As for "clear steep barriers" - again, just clueless stuff. There weren't clear steep barriers to search when Google started, there were dozens of search engines. Google created them. Creating barriers to entry is expensive and the "FCFF people" imagine they arrive out of thin air. It takes a lot of time and or money to create them.
It's unclear if "the model business" is going to be high or low margin. It's unclear how high the barriers to entry for making models will be in practice. It's unclear what the reinvestment required will be. We are a few years into it. About the only thing that is clear is this: if you try to run a positive free cashflow business in this space over the next few years, you'll be crushed. If you want a shot at a large, high return on capital business come 2035, you better be willing to spend up now.
There is not a single thing in there which is counter to valuing a stock. Anyone with half a brain knows large free cash in the future can more than make up for negative free cash flow in the early years. You can't point to a single thing, you are just hand waving.
Fanless x86 desktops are a thing too, in the form of thin clients and small PCs intended for business use. I have a few HP T630s I use as servers (I have used them as desktop PCs too, but my tab-hoarding habit makes them throttle a bit too much for my use - they'd be fine for a lot of people).
My experience with fanless Intel is that they tend to be rather sluggish for desktop GUI use, though. Which doesn't seem to be an issue with Macbook Air.
I've been experimenting with Asahi Linux recently on a spare M2 Air I have lying around, honestly very impressed. It's come on a lot since I last tried it a year or so ago
> I don't think you could built a fan-less x86 laptop.
Sure you can, they’re readily available on the market, though not especially common.
But even performance laptops can often be run without spinning their fans up at all. Right now, the ambient temperature where I live is around 28°, and my four-year-old Ryzen 5800HS laptop hasn’t used its fan all day, though for a lot of that time it will have been helped by a ceiling fan. But even away from a fan for the last half hour, it sits in my lap only warm, not hot. It’s easy enough to give it a load it’ll need to spin the fan up for, but you can also limit it so it will never need its fan. (In summer when the ambient temperature is 10°C higher every day, you’ll want to use its fan even when idling, and it’ll be hard to convince it not to spin them up.)
x86-64 devices that are don’t even have fans won’t ever have such powerful CPUs, and historically have always been very underpowered. Like only 60% of my 5800HS’s single-threaded benchmarking and only 20% of its multithreaded. But at under 20% of the peak power consumption.
Sure, I have one sitting on my desk right now. It uses an Intel Core m3, and it's 7.5 years old, so it can't exactly be described as high performance, but it has a fantastic 3200x1800 screen and 8GB of RAM, and since I do all my number-crunching on remote servers it has been absolutely perfect. Unfortunately, the 7.5-year-old battery no longer lasts the whole day (it'll do more like 2 hours, or 1 hour running Zoom/Teams). It has a nice rigid all-metal construction and no fan. I'm looking around for a replacement but not finding much that makes sense.
It can consume almost 20W sustained, which is quite a lot. Competitors will definitely have fans roaring at this power draw. I think the all metal design makes a huge difference from a cooling perspective. The entire case is basically a heatsink.
You can, the thing is you have to build it out of a solid piece of metal. Either that's patented by Apple or it is too expensive for x86 system builders.
If I recall correctly Apple had to buy enormous numbers of CNC machines in order to build laptops that way. It was considered insane by the industry at the time.
Apple, unlike a lot, if not all large companies (who are run by MBA beancounter morons), holds insanely large amounts of cash. That is how they can go and buy up entire markets of vendors - CNC mills, TSMC's entire production capacity for a year or two, specialized drills, god knows what else.
They effectively price out all potential competitors at once for years at a time. Even if Microsoft or Samsung would want to compete with Apple and make their own full aluminium cases, LED microdots or whatever - they could not because Apple bought exclusivity rights to the machines necessary.
Of course, there's nothing stopping Microsoft or Samsung to do the same in theory... the problem these companies have is that building the war chest necessary would drag down their stonk price way too much.
They just don’t want to bet they can deploy it successfully in the hardware market to compete with Apple, so they focus on other things (cloud services, ads, media, etc).
Google is not a hardware company (outside of the Pixel lineup where they just take some white-label ODM design).
Microsoft has a bit more hardware sales exposure from its consoles, but not for PCs. They don't have a need for revolutionary "it looks cool" stuff that Apple has.
Amazon, same thing. They brand their own products as the cheap baseline, again no need.
And Meta, all they do is VR stuff. And they did invest(ed?) tons of money into that.
The point is they have enough cash to make an attempt to be whatever company they want. Apple chose to delve into hardware, the others chose not to, not because they don’t have the cash.
Do people not use their own government as the entry point for visa applications? I just go to the website of the foreign office of my government which has a list of the requirements to enter every country in the world and what needs to be done and how. I have never started with a google search.
