In Google's account system, there's ~nothing you can do with just the username. You need a user id. The only way you can get a user id from a username is to call an RPC service to do the lookup. Doesn't matter whether it's for logging in, finding the account to deliver email to, creating an account, or whatever. Everything goes through that chokepoint. And that RPC server will always normalize the username at lookup time.

(I probably read that normalization code more often than anyone when working at Google on account security stuff, since that was the single place that had any reason to deal with usernames rather than user ids. That code was effectively immutable.)

No, I already explained why rebutting the whole article is impractical. I might have been ok doing so for a single argument just as a demonstration, but your unwillingness to point out the most compelling and impactful argument of this article makes it pretty clear there was no point to it. It was 10k words of padding.

> or his main point about profitability

A message ago the main point was about revenue growth. Now it has shifted to profitability. So, no, I'm not going to waste time rebutting an ill-defined, second-hand, moving target argument either.

> I don’t imagine calling his theory “obviously nonsense” was as far as your reasoning went. Care to expand on it?

It's nonsense by definition? The claim is that companies have too low a revenue to pay for their compute spending commitments, and growing super fast can't fix that problem. Super fast revenue growth is exactly what solves the problem of revenue being too low! It's a very strange thing to assert, but it's what he did assert, and it's the only bit of this article that's even close to what you claim is his main point.

("Oh, but super fast growth can't happen because of blah blah blah as explained in some other article", you'll say. Maybe so. But blah blah blah is not what Zitron wrote in this article, in this one he wrote that revenue growth won't solve the problem of low revenue. Rebutting a 10k word ramble is merely impractical. Rebutting an amorphous blob of 1M contradictory words is impossible, there's always going to be some explanation about how what he really meant is somewhere else.)

The sleight of hand (yours) is that those numbers don't seem to be from the article. Like, none of those numbers appears in the article except the years, and the years are used in different contexts.

Likewise what you say is the main point ("the labs can't grow fast enough to hit these targets") is not something the article makes an argument about!

In fact, the only even related thing in the article is an this assertion:

> To be clear, “growing super fast” is no longer sufficient for OpenAI and Anthropic.

Which is obviously nonsense, but also at cross-purposes with what you say.

So neither what you say is the main point nor the numbers is in the article at all. Seems pretty sad as demonstration of how compelling the article was.

I've seen people talk about how compelling they find these articles, but literally never have they been able to point at a good argument. Like, in a paragraph, what's the most compelling and impactful argument in this article? What are a couple of numbers that make it so?

https://blog.arxiv.org/2023/01/31/arxiv-announces-new-policy...

For some reason comments got moved to this one.

(Just a couple of weeks ago there was an equally implausible frontpage post showing the Firefox had more than doubled in use in the US, to 12%, but only in the US, not in any other country. My money is on scraper bots identifying as Linux-Firefox, run from US IPs to minimize the impact of geo-blocking. But when Statcounter doesn't explain their methodology and doesn't care about the quality of their data, we'll never know.)

There's a bunch of tiny Google products like that. Google Reader is a good example of this -- the reason the shutdown didn't lead to a cambrian explosion of RSS readers is that there wasn't a business to be had there, even a small one. The new AI-first Google Finance UX makes me really sad, but it's 100% certain that nobody else will be signing up to run a clean, free, adless finance site. Books, Scholar, Patents are probably under a lot of pressure, and if they close, don't expect anyone else to pick up those golden opportunities.

Edit: Ok, reading throught the rest of the comments, it looks like the article was written by a muppet and not actually about Google News despite the title saying "Google News". But I guess the above still stands. The golden opportunities aren't there.

(Each reader can decide themselves whether what you wrote is unfair to Zitron or to LLMs.)

But "had" still doesn't mean what you are implying: once the model solved the problems and the solutions were verified, the problems weren't open any more, so a later description using the past tense is totally consistent.

I doubt any credible analyst has claimed that. What's the total AI capex that's already been spent? About $1T? It's a pretty absurd idea that those DCs need to make $2T/year for 5-7 years -> $10T-14T over their lifetime to break even.

(Yes, this is nitpicking in the sense that there are probably analysts talking about how projected and sustained capex at >1T/year will require 2T/year in revenue, so patching the article won't be a biggie. But this article is cosplaying as financial analysis and leads off with such an obviously incorrect argument. What does that say about the credibility of the rest of the article?)

> Diseconomies of Scale.

Another very basic mistake here. The author starts talking about efficiency in the context of past technologies. That's lower unit costs as scale increases.

But for AI, they seem to switch from talking about unit costs to total costs. Or at least I can't explain what they say about models getting more expensive over time in any other way, because that is not true about unit costs.

We've never seen economies of scale as large as for AI. For a given quality level, the cost has been dropping at >10x per year, not increasing.

Ok, so what is the botec? They don't say, but it's probably something like "next year's $1.5T/year in capex * 50% return on invested capital + a little bit extra for opex".

Obviously a given year's capex doesn't need to return its investment immediately the following year, but over it's ~5-7 year depreciation period. So in making this botec, they're not just taking next year's capex, but extrapolating it to a steady state of $1.5T/year in capex.

But why would that level of spending be steady? It's exceedingly unlikely to be. Future capex is not locked in. It's contingent on revenue and revenue growth. AI revenue has been growing faster this year than even the most optimistic projections suggested. It's hardly a surprise that capex projections were dialed up. If revenue growth lagged instead, it would go the other way.

(You need only look at Google Cloud growth and margins to get an idea of how good an investment last year's AI capex was. But at the time, the arguments for it being crazy were identical to those made today, except with the numbers substituted.)

There's a separate issue, which is that you can't really think about this on a sector-wide basis. In most businesses there will be winners and losers, demanding everyone be a winner is unrealistic. E.g. right now anyone with the business model of renting out the compute is being showered in money, people with a business model of building non-frontier models are losing money. The latter group's bad strategy doesn't invalidate the former group's good business.

Like, there's three plausible arguments about safety of open models:

1. Any concerns are fake news. Open models will always be safe.

2. Safety is irrelevant. Open models should not be regulated even if they're unsafe.

3. Safety is a technical problem with technical solutions. People releasing open models should invent and implement such solutions.

I think option 1 is totally out of touch with reality.

Option 2 is at least self-consistent, it's the argument being made by people who will say that all regulation is always bad. It's also like the worst possible world from an x-risk perspective (but I realize that the average HN poster believes any x-risk concerns are just frontier lab marketing).

Option 3 is playing on hard mode compared to proprietary models, which can both implement additional safeguards out-of-model and prevent modifications of the model. But if the answer to it is "it's too hard, Anthropic needs to come up with the technical solution", then that's not exactly a ringing endorsement for the safety practices of the open model labs, right?

Just this article has a dozen instances of numbers being used to prop up the argument. They're mostly irrelevant to whatever argument Zitron is trying to make at the time. Some others are straight up misrepresentations. HN really is not a good forum for point by point rebuttals for that many cases. Do you think there's a particularly high impact use of numbers in this article?

[0] https://x.com/binarybits/status/2034376359909130249