dofm

Megaspeed is a "Singapore-washing" operation. It builds data centres that serve the Chinese market specifically.

The results are ultimately Chinese data centres no matter where they are (Malaysia and Indonesia apparently).

But Megaspeed's parent company/backers build data centres in China and there is not enough (if you'll pardon the pun) of a Chinese wall between the things the parent builds and the subsidiary builds; they apparently shared the same promotional material.

All of this is stuff Jensen Huang does know. (Not saying you are doing this but) I think it's hilarious anyone is out there allowing for him to have an innocent "I was preyed on and could not have known" excuse. It doesn't stack up. He knows exactly what Megaspeed is.

FWIW, I guess it probably doesn't always matter too much in a tool where you want a logical, correct, standard, robust outcome like agentic coding.

Those outcomes are more easily averaged and we are after all talking about artificial intelligence. And for a coding agent… perhaps this is the right tradeoff.

I just think it's important to note, every now and then, that words aren't feelings or thoughts, and that an author's next book cannot be calculated from only the words of their previous books, etc. Basically we should understand that the reasons why we can't generalise from "it can design processors, write code and solve maths" to "so human-expert-quality writing and web design should be easy enough" are probably rooted in differences like these.

It contains the averaged written ghosts of the outcomes of those things (as well as, I am sure, some descriptions of those experiences, attached to descriptions of the outcomes — "so I was sitting on the bench on the common and…" would be routine in a training set derived from me!)

But they are not the things themselves — they are incomplete representations; just words that approximately describe the application of that intuition to the problem at hand.

Think about how you visualise complex code projects in your mind — how it's not quite diagrams or buildings or words but multiple things in a sort of multi-dimensional, itchy space that is satisfied when a really good implementation appears. Think about how often the code you have built retains that wordless shape long after you wrote it down. Think about how impossible it is to adequately describe it.

An LLM can't have intuition itself.

It can only simulate the written outcomes of intuition in its training set.

For sure, some of those written outcomes are a good guide to reaching the outcomes of intuition.

But the likely uncloseable gap is that some intution is unspoken, literally felt by the brain's default mode network and by parts of the prefrontal cortex.

LLMs basically never shower, do the washing up, kick their foot in the dirt while sitting on a park bench with a coffee: the domain of the default mode network. And they never sit frustrated looking at things, guided by their long experience, with the itchy brain feeling that there is a solution before they can even say the words.

A lot of this stuff happens to us deep down, way outside of the reach of words. And as a result, some aspects of intuitions are just never described in words.

s/could not have been compromised/could have been compromised/

And a more powerful and flexible device than the original Pi — the Zero 2W — is cheaper.

You think Jensen is sort of mostly a victim rather than very obviously negligent? That he couldn't have known a Singapore-based Chinese woman whose company builds Chinese data centres could not be compromised and acting on behalf of the Chinese state?

I mean I think many things about Nvidia but one thing I do not think is that its CEO is an idiot.

I also think he will have had security advice that she might be, not least because Trump and Rubio knew a year ago.

Here is an easy guide to working out if someone could be a spy: if they are foreign, keen to get to know you, and they run a large enterprise that would be materially advantaged by having a way around the export controls on your products that impact the state where they were born and where they do business, they could be a spy.

Only if they gain that ergonomic advantage without complicating flow of control. Metaprogramming is costly for example, in the past so have been operator overloading and lazy evaluation.

Is this not 1.875% of Nvidia or something?

I don’t think the divorce is the real story here.

s/Postgresql/Postman/ obviously.

Brain concept leakage there. I meant all those apps that start off with a simple purpose and end up with an enterprise cloud plan.

I guess the people who downvoted this must enjoy egg but hate eating it off plates:

https://wccftech.com/nvidia-intervened-federal-investigation...

> dog's sniffing class

There's a blog post in this for sure. And a classic joke.

I agree.

For me, and for customers I want to document things for, it has come down to MySQL Workbench on the free/open source side of things.

I want to support multiple platforms without having to fundamentally document multiple strategies.

Most DB tools feel like they will go the way of PostgreSQL or Insomnia — get slowly burdened with cloud facilities and then a bait-and-switch means you need a login somewhere.

I would much rather be able to talk about clients that do fewer tasks but do them well, but DB clients do seem to be prime enshittification territory for some reason.

> Every time a new model drops it feels wildly powerful, then regresses to some sort of generic SOTA level we've been used to over the last 6 months.

Is this 2026's "my macbook feels 'snappy' when I update it", though? (That is to say, not really much in evidence)

House of El covered this in her latest video incidentally.

Not really anymore, no.

The Foundation and the Trading Company have different goals, and the Foundation is not wedded to the Linux SBC Pi at all for this.

Physical computing education with Python is more than adequately served by the RP2040 and RP2350, and that is more or less where their focus is.

(Beyond that I think most of this is just geek whining, to be honest.)

Yeah, currently I have a 3B, 2B and an original wifi-free zero in a drawer.

The 3B was my Octoprint server for the longest time while I was testing the 4B for a product, but for reasons I don't really understand, the 3B isn't quite punchy enough for Klipper, when it fully should be; it drops connections to the printer. I think any of them may continue to have some uses but perhaps not till I move to a new place.

Haven't gone for a 5B or Zero 2W yet, don't really need to. Rather more interested in what happens with the Picos these days.

Oh this is good news — I have been looking for a fuss-free cross-platform alternative to MySQL Workbench to recommend for a data product for some time.

(Workbench is a pain but it was capable, off-the-shelf and was possible to document it once, more or less, for multiple platforms; I did not actually hate using it. But alas it is old now)

I will surely test this out; good luck with it.

Is it a worse misallocation, I wonder, than doing what most people do with Raspberry Pis, which is leave it in a drawer ;-)

I don't agree with him on everything but one of the questions Ed Zitron has raised is — have all the GPUs Nvidia has sold actually been manufactured and warehoused?

If they haven't been manufactured, things are potentially quite sketchy.

Both the market for compute and the stock market valuations of AI firms (and theoretical valuations) are overcooked.

The problem is that "scarcity of GPUs and RAM" is being used as a proxy for "demand for AI".

It's a broken proxy; the scarcity of H100s is artificial.

Is there evidence they are unambiguously making profit off inference when all the supply costs of that inference are considered? It does not seem likely.

I guess we will see. But I would not expect GPU hardware to last that long until there is some underutilisation in data centres.

Mind you, most of the H100s that have ever been sold are still in their packaging, I suspect. Or at least not plugged in.

> At current prices ? those GPUs that are there now will be running there in 10 years still.

If they don't burn out after two or three. Annualized failure rates are about 9% according to Meta.

A quarter will have failed within three years. Less than half of the original hardware will be running after ten years. In both cases if you assume that rate does not increase over time. Do H100s come with replacement guarantees?

They are in Nvidia warehouses. Nvidia is selling H100s and other hardware in advance to firms who do not have data centres to put them in.

Nobody is getting fired for thinking they need to buy scarce H100s in case they aren't able to get them in the future. Nobody is even getting fired for keeping their companies out of Jensen Jail — the fear that if they don't buy the whole server stack from Nvidia they might not get future GPU allocations.

There really are analysts who think Nvidia is channel-stuffing.

If the market deflates there will be a glut of very cheap compute, both in terms of data centres (though many of them will be able to offer managed model training services instead, I suspect) and in terms of unused hardware being made available for sale.

Nvidia won't lose out; everyone else is going to.

Offered isn’t the standard, is it? Were his devices hacked? Was he bugged? Has he been compromised in other ways that might make him a further blackmail target?

Plus — who else knew is a problem. Apparently Trump knew! Do we think he never used that?

If this was any other high value C suite employee, Nvidia would fire them for negligence.

And that is before you look at the mundane — did he ask any subordinate to lie to his wife? In my book that's a firing offence.

Training is not infrastructure. It’s not capex, not “investment”. It’s really an ordinary running cost of what they do, for the foreseeable future, and it’s absolutely enormous. They couldn't even dial back training in a winner-takes-it-all scenario, because training of smaller, domain specific models would remain a way to compete.

It could be absolutely exasperating, even in 2010 (especially then since browser support for touch events was still shaking out).

The problem is that even without a framework, you didn’t have a blank slate (which would have been better). You had a moving target of blank slates, you had platform differences, you had drag vs touch-drag, entrenched older browsers and the simple frustration of testing on an array of real devices, all sorts.

Now, if you were talking about 2015, then it starts to get a bit flatter.

I think this is correct except that it ignores the hidden bubble in the SPVs: a lot of the firms who look only partially at risk have simply moved debt off their balance sheets- but to whom? Pension funds, sovereign wealth funds etc.

And “OpenAI bubble” alone has one massive implication: the likely demise of Oracle, a blue chip business that is deeply embedded into the western world.

On the one hand, if the air is let out of the bubble before those SPVs get themselves too deep into building things that won’t be needed, that is good. But nevertheless the failures of the SPVs will harm people who are not in the exclusive billionaire club.

Also there is a considerable money-go-round; the money for everything will dry up fast when the cloud LLM businesses can’t convince people to give them money to buy Nvidia GPUs anymore l

To what extent do people “on the street” as it were relate this to the real estate bubble?

Did you ever actually do this work?

Do you remember what browser testing was like?

Remember the quirks of browsers’ different understanding of transparency, of z-index, of events, of DOM compliance?

I think the suggestion that getting it right could take a week in the context of a particular application is actually not particularly hyperbolic.

Mind you, I was doing cross-browser testing long enough ago that I am one of those people who can say they independently invented AJAX in-house.