josephg

You also need to look at memory size (to store the model). And processing can still be the bottleneck.

Yeah. It would require LLMs to sweep through other professions like they have in computing. I don't see strong evidence of that happening yet.

I’ve spent more money on Claude this year than I’ve given to a lot of other tech companies, combined. Well, I bought a new phone too - but amortised over the phone’s lifetime it’s still much cheaper than I’ve spent on my claude subscription and what I’ve gotten my company to pay in api usage credits.

There are two questions. Will the value keep scaling? And how much of that value chain will anthropic hold on to? Right now software development is being eaten alive by LLMs. Will that happen to other industries too? And in the long run, what do the models we run look like? Are people going to end up self hosing opensource models themselves, or are we going to collectively hitch our wagons to frontier models provided by anthropic and their competitors?

If ai keeps getting better - which I think it will - and if anthropic can corner a large part of the market in an enduring way - which seems less likely to me - then yes. I can easily imagine them becoming a huge company.

> unless you choose to make data from these services available to other Google services that support personalized ads

I don't know if android has changed in the last 10 years, but my last android phone had constant popups in official google apps begging me to enable all sorts of information sharing right across google's ecosystem. The popups were constant. And it was full of dark patterns making it difficult to say no.

Google probably assumes you only see this stuff once. But if you say no, some features didn't work properly and it keeps bugging you.

I don't know how or why, but I wouldn't be at all surprised if 90% of gmail users unintentionally opted themselves in to this "choice" before the end of the year.

> I think the idea that CAs are uniquely incompetent is probably incorrect.

The problem with the certificate authority system is that any CA can sign a cert for any domain name. It only takes 1 incompetent or malicious CA to break the whole system wide open. And there are so many CAs that inevitably some of them are incompetent, and use keys like 7. And some are definitely in bed with a government you don't like.

Moving certs into DNS means we only have to trust the registrars. Not the registrars AND the huge set of CAs.

> until they comply so maliciously you'll need a new term for that.

The ruling is pretty clear that malicious compliance won't be accepted. There's a whole section on all the things apple isn't allowed to do to get around this ruling, starting with this:

> the first respondent must not (including by procuring any act of any related corporation) prohibit, prevent or take any action (including by imposing a fee or condition) which is likely to hinder users of iOS devices from discovering, obtaining and updating iOS apps, including alternative app stores, via Alternative Distribution Channels, or from using iOS apps obtained in that way on iOS devices, or prohibit, prevent or take any action (including by imposing a fee or condition) which is likely to hinder app developers from distributing and updating such iOS apps to such users via Alternative Distribution Channels, ...

Read the ruling. It's clearly written with Apple's EU schenanagens in mind:

> 4. Subject to orders 12 and 13, the first respondent must not (including by procuring any act of any related corporation) prohibit, prevent or take any action (including by imposing a fee or condition) which is likely to hinder users of iOS devices from discovering, obtaining and updating iOS apps, including alternative app stores, via Alternative Distribution Channels, or from using iOS apps obtained in that way on iOS devices, or prohibit, prevent or take any action (including by imposing a fee or condition) which is likely to hinder app developers from distributing and updating such iOS apps to such users via Alternative Distribution Channels, including but not limited to: ..

Then it goes on to list all the ways Apple isn't allowed to quietly break 3rd party app stores - like, they can't make developers or users pay money to use them. They can't restrict which APIs are allowed to be used via 3rd party stores. They can't require that people who publish apps in the official store remove their apps from 3rd party stores. And so on.

Better to trust 1 organisation than trust 100.

> (Moving the PKI into DNS does not remove it, it just moves it around.)

Better in DNS though. You could have certs mirror the structure of DNS. Install a single root signing key for the top level DNS registry on client devices. Then you can query that to find out the signing keys for .com, .net, .au, etc. And continue from there.

One big problem with the CA system is that any CA - even the most incompetent - can sign certificates for any domain. If trust was moved to DNS, there would only be a single organisation that could forge certificates for any given domain. Using 7 as a private key? What a joke. The CAs aren't worthy of trust.

We already have to trust the domain name registrars anyway, since CAs query them to figure out if you own a domain before signing a certificate. In total, removing CAs would not only simplify the system. But it would allow scoped signatures. And we'd end up needing to trust far fewer actors to have a secure internet.

A secure DNS is valuable whether or not we get rid of CAs, since CAs rely on DNS queries to verify domain ownership.

And if we have a secure DNS, then CAs are completely redundant. I'd much rather trust a handful of domain registrars than a handful of domain registrars, plus 100 random questionably competent organisations all over the world. Especially when any one of them can generate keys for every domain on the planet.

He's right. DNSSEC is a mess, and we should fix that. We should fix that whether or not we delete CAs because CAs rely on DNS to verify that you own your domain. If DNS isn't secure, neither are CAs.

In the long run, fixing bugs is almost always better than working around them. DNSSEC is a bug. So are CAs. We can and should fix both.

Make dns secure and put certs / public keys in dns. Control over dns already gives you the right to get tls certs. It should just be connected directly. We don’t need CAs.

Unfortunately this isn’t possible. If we spent billions upgrading electricity infrastructure, sure. But it’s much cheaper overall to make smart use of solar panels and batteries that are already being installed in the grid. And that requires a smart grid, full of networked equipment.

For example, during peak sunlight times, many neighbourhoods now generate more solar than they use. Without a sink for that power, voltages would go haywire on power lines and damage equipment. So the grid uses control boxes to manage how much power each house is allowed to send back to the grid. And all that needs compute, live metering and so on.

I’m not surprised! The amount of wasted memory and compute in modern software is astounding.

If you’re a developer, try promoting Claude to profile and optimise a piece of some piece of software. I routinely see 20-40x improvements with a single prompt and a few hours of opus credits.

Thanks for your work by the way. At a large scale, it is much more efficient for some software developers to do a bit of belt tightening in their code than ask millions of customers to invest in bigger computers. Globally the savings are enormous.

What savings?

Disconnecting is also effective. Fridges don’t need to be on the internet.

It’s the same in a lot of sectors. Airlines. Electricity companies. Probably train lines. Banks. Fire services.

With all the disruption from AI on our industry, I can’t help but feel a little jealous.

> Unless you are working at a 'tech' company, the norm is that even developer machines will be constrained on memory usage for a given workload

Honestly I think most of us are just spoiled kids crying when our expensive toys are being taken away. If memory cost 10x more to manufacture than it does now, we’d just have 10x less ram in our computers and our software would be written to deal with that. We’ve done it before. My first computer had 4mb of ram and it ran windows 3.1 just fine. We still had working compilers. If ram were 10x cheaper before the supply shock, I think we’d be crying that companies won’t buy us 320gb ram in our laptops. The java and rust compilers would somehow be even less ram efficient. Firefox and chrome would still happily use it all to run adware across the same 50 tabs I want to keep open.

Theres good reason for LLMs to need that much ram. But very little else we do on our computers needs anywhere near as much ram as we use. Have faster computers made video games more fun in the last two decades? I don’t think so.

The supply crisis is annoying. But I’m also a little glad there’s a reason to push back on fat, lazy software.

You say that. And yet, most of mad max: fury road was filmed practically. It looks amazing.

I could easily imagine the right films renting these cars for a stunt. Film cameras and practical effects are having a resurgence.

On Linux I’ve taken to disabling swap before doing memory heavy operations. Claude will carelessly run benchmarks and tasks which gobble up all available memory. When that happens, I’m not sure why but my system becomes unresponsive and I can’t kill the process. If I disable swap, the oom killer takes care of whatever Claude was doing and Claude figures it out after the fact.

Those problems can also be solved with a better organised electoral commission. Here in Australia we sometimes have long lines at polling stations, but I’ve never heard of any of those other issues occurring.

This is one big advantage of paper vote systems. You can still bribe, trick or coerce a single person. But it becomes much more difficult to do so en masse. Computer voting allows fraud at scale. It is much more difficult to scale an attack on a paper voting system.

> recently, they removed the ability for citizen to stay during the counting

Here in Australia, representatives of the various political parties observe the counting as scrutineers. I did it once. It’s absolutely fascinating seeing the system and rules. All (lower house) votes are physical slips of paper. Officials sort them into piles of 50 based on their first preference, in a big warehouse. Scrutineers watch them do it. We are allowed to contest their reading and sorting of a vote, but not physically touch anything.

I’m sorry to hear you can’t see your own democracy in action. I have more trust in our election process seeing it up close, in person.

> They will never sell a car with similar functionality to that demo to anyone

They're explicitly advertising selling exactly that car. They just say it's by "invitation only". Do you think it's just a marketing gimmick? I dunno who's crazy enough to buy one. But for the right price, they could definitely sell a few.

Depends how much they sell them for. It could probably make the investment back if they price it at $5M or something and sell 100 of them to the ultra rich.

I sometimes ask an llm to plan out an algorithm or concept in Haskell. Then I ask it to implement it in rust. I think I get a better implementation that way - though I might just be fooling myself. They seem to find simpler abstractions if you start there.

I used to use GetRight to download big files over FTP or HTTP, over dialup. You could set it up to automatically re-dial your ISP if you got disconnected halfway through a download. And you could download a file in multiple "threads" - for example, get 2 concurrent HTTP sessions downloading the same file, one thread starting from the start and a second thread starting halfway through the file.

GetRight did all this in 1997.

You can do actual html. This is Leptos, a Solidjs like rust library:

    #[component]
    fn App() -> impl IntoView {
        let (count, set_count) = signal(0);
        view! {
            <button
                on:click=move |_| set_count.set(3) >
                "Click me”</button>
            <p>{move || count.get() * 2}</p>
        }
    }

I don’t think we ever tried to make wave a social network.

> everyone is doing it

Not in my house. What is even the point of connecting a coffee machine or a washing machine to the internet? I think my washing machine advertised that I could download new washing cycle programs in the app. Who on earth cares?