hedora

And residential proxies are becoming increasingly legitimate because of over-zealous AI blocks.

They should be held liable, in the same way as brick and mortar stores are liable if they sell you shoddy goods, and the manufacturer can’t be found after your house burns down or whatever.

I need to mess with my local DNS resolver for unrelated reasons. Is there a way, in 2026, to setup DNS in a federated way?

I could point to ISP DNS, or 1.1.1.1 / 8.8.8.8 / etc but those are all centralized, and are all subject to surveillance and censorship laws.

DNSSEC, etc, are something, but they rely on CAs. I mostly just want some sort of system that's outside centralized control, but that can be externally audited for trustworthiness.

We have a lightning specifically because we need a full sized truck. Our RAM ICE's cargo capacity was barely over 1000 lbs (BMW i3: 800 lbs). The Ford EV's capacity is 2250lbs.

Also, the RAM's transmission overheated on gentle slopes.

The slate looks great (more cargo weight capacity than our full sized RAM ICE truck; half the price, much lower fuel costs, much more convenient to own).

If their second model is 250 class, we'd probably buy it. The Lightning can only tow 10,000lbs, so it's not great at hauling rock, etc -- it can do it, but only ~2.5 tons per load when you add in the weight of the trailer. This means we have to keep paying $100's to delivery services.

It'd be nice if that hypothetical 250 slate seated 5-6 people too. That'd let it be a backup kid hauler, like the lightning is. The lightning has more bells and whistles, but the aesthetics of the slate are much better, and I like the idea of just buying an aftermarket radio that does not integrate in with the power train.

Of course, if we didn't need a truck, we'd get a Rivian R1S/T. They're great for camping, etc, and what most people want when they buy a pickup truck as a family car.

The Lightning didn't really fail. It was the best selling EV pickup in the US by a large margin.

I think it was killed for political reasons: They repurposed the plant for AI datacenter batteries. I predict we'll have EV pickup trucks again if the 2028 election looks like the 2026 election.

Anyway, my next full sized pickup will also be a pure EV. If I can't get a Ford, then I'll get something else. (The lightning is far better at truck stuff than our late model GMC and Ram trucks were. Maybe I can get a truck from China + register in Mexico? Is that a thing?)

The new Lightning has a useless serial hybrid setup and gets 700 miles of range (who needs that?). I suspect it'll end up being like other serial hybrids, and an option to omit the gas engine will be available, and also substantially lower the price (or enable a larger battery). In the meantime, they'll tell Trump that he managed to kill EV pickup trucks.

Permitting in California more than doubles the installation cost of home batteries + PG&E gets to screw with you if you buy one.

Propane generators here have the same problem.

That still doesn't explain the price gap. You can literally just get a V2H EV and use it instead of a home battery (or, you could if the ecosystem was more mature).

In our area, when a storm hit, the roads washed out, so the generators ran out of propane. Then the local propane companies ran out of delivery truck capacity because of the demand surge. Finally, the regional distributors ran out of propane.

Our solar panels kept the fridge running, but we were heating the house with a wood stove (it was very cloudy, so we were getting maybe an hour of sun a day). If we had bidirectional house charging, we could have had warm showers + used the heat pump: We would have been able to fast-charge the car at 100-200kw in town. The house can limp through 20-30 hours on ~ 25kwh, so the charger would charge the house at a few days per hour, including losses from driving to and from town + round tripping through the battery, and also running the house heat pump.

None of the other people around here stayed online. Some propane-dependent people had no electricity for > 2 weeks (with rationing, they managed to keep their generators online for about a week, and PG&E took three weeks to fix their lines).

This was in the outskirts of Silicon Valley, so better-than-puerto-rico infrastructure, but only just.

It is difficult to get to M * N * P control flow edges with O(1) result unwrapping code.

In java, the caller gets the reference before the constructor runs. It's in the language spec. It's terrible. You can prove it yourself with something like:

   this.dangerous = new Lol();

and then have Lol()'s constructor sleep for 10 sec then set done = true at the end of the contstructor. In another thread:

   while(!this.dangerous) { }
   assert(this.done) // will throw.

I guess you could use double-checked locking as a class author by having every public method on your class do a if (!this.done) { throw; }, and make sure there are no public fields.

That's pretty lame though.

As the author of the class, I cannot prevent my caller from leaking the reference, so my public API allows for pre-constructor invocations of all my methods.

Software transactional memory doesn't really work with exceptions, since it precludes the use of catch + finally. It looks more like a top-level panic handler (where the throwing code cannot permute the heap, which is cool). To get exceptions, you'd need nested software transactional memory, which gets into obscure transaction processing theory. There's also the problem that your transaction could side-effect the outside world, then throw and be aborted, so you need to rework how I/O works if you walk down that path.

Exceptions are basically write-only, so I agree that their ergonomics for that use case are pretty good. Panic is better though, since it's really no error handling. Panic is easier to test, and easier to read / understand.

Not installed universally, and it also doesn't reliably run python programs. (Minor + major semver changes break backwards compatibility, and pip mutates the global system state.)

You can always write:

   x : Result<string, err> = function...
   dangerous_function();
   switch(x) { 
     Ok() ... 
     Error() ...
   }

There's no difference between result and null other than readability (there's a mechanical transformation between the two, where you use unwrap() in rust on an unchecked null dereference, and you read errno in C instead of looking inside the Err() result variant).

To be clear, I strongly prefer Result<> because of the readability, but any dumb thing you can write with NULL + errno can be written with Result (and vice versa).

How do you handle out of stack space on 'authenticate_user'? What about the machine check exception loading the DRAM page that contains "successfully authenticated"?

I'm drawing the line into the past. Back when Biden was president, a computer that currently runs Flash Next 125B well cost $1400.

But yeah, I assume that when the economy gets back to normal-ish, in 2024-equivalent dollars, we'll be at 500B-1T for < $2000. That'll be 3-4 Moore's Law doublings since computer prices were sane. 125B * 2 ^ 4 = 2T.

Yeah; the open model markup goes to zero. The manufacturers of the "good enough" inference chips are probably going to be some subset of NVidia, AMD, Apple, smartphone SoC manufacturers, and (long shot) Intel.

I shortened my list to NVidia and AMD because putting Apple in muddies my point: I'd argue they'll be able to sell laptops + phones in that world, but only if you can run frontier open models on their devices, and do so efficiently. This is definitely not the case for their in-house AI offerings, like Siri, but it is already true if you install llama.cpp or whatever. I suspect that AI will not be a big revenue source for them (unless they completely turn around their software engineering org), but instead will be a "good enough to keep people from dumping their iCloud subscription" offering instead. Siri is basically that: It's completely useless, but people aren't switching to android because of it.

7M is fine if you don't care about the quality/correctness of the output and just want some utterance that looks plausibly related to the question. I guess that is the bar this thread has set for Google search quality being good enough for most users.

I think you can press a button + say "Hey Siri" at this point (though that assumes Siri works at all, which I agree is a stretch.)

I have literally never had that experience with any of the following search engines:

- DDG

- Kagi

- Bing (10+ years ago)

- Yahoo (I'm getting old...)

I searched by typing lists of keywords into the search box in the past. Maybe you're one of those people that types sentences, or relies heavily on synonym matching? (Which finally works now, but only if you turn on LLM mode -- even then, Google shipped that months or years after kagi, and the reviews are scathing vs. the performance I get out of the Kagi AI search.)

Come to think of it, my Qwen box has gone up in value at a rate of $235 / month since I bought it. I guess I could sell it + it would retroactively pay for my Anthropic subscription.

Apple is collaborating with Trump. For example, they helped demolish a third of the White House. Before that, there was the #AppleToo thing, and nothing happened.

Using taxpayer money and kickbacks to ram through data center construction is bad, but it's not even top ten on the list of crimes the executive branch committed in the last two years.

If the core function of the university library's archival system is not providing free access to their collection, then what is its core function?

But, per groups of queries, the worst they've seen is 11 thousand, and that's across many libraries. So, some tiny fraction of their users amplified their traffic by 11,000x by using a misconfigured LLM. I've done that a few times without LLMs (who hasn't passed --recursive to wget without remembering the -np?)

Also, if they just turned off all the user-hostile crap, they'd probably get at least 1000x back on their serving costs. How much does it cost to serve a login gate, captcha and rejection screen vs. a ~1MB static file?. What about the human cost where actual users cannot get access to the data they should have access to?

Just now, I ran a Kagi search. Its LLM summarized some page on the internet, but when I went to check the reference, I got a "sorry, humans only" access denied error message. I guess it's good that I could ask the LLM to quote the document verbatim for me in a follow up message, but what fresh hell is this? (Also, the error message screen was clearly longer than the document I'd requested.)

I was not aware of that (terrible) court decision. Wow.

I suppose this is one of those laws that doesn't apply if an individual is a victim? I have a email box that sometimes fills with spam and rejects legitimate email. Can I get the corporations that filled it up to be charged with felonies?

11 THOUSAND searches in one day? What's the problem?

If anything, libraries should be becoming more AI friendly, not less. Instead, we get this sort of bullshit from the ACM:

https://dl.acm.org/doi/10.1145/3832783.3834511

It's a canonical DOI on what was a download server. It says exactly which file is being requested. (I chose a random recent one.) At the very most, they should serve a static page with a bibtex entry, and a few other metadata fields, and a link to the PDF and whatever other formats are available.

Instead, we get piles of captchas, login-gate spam about how they're "open access" now and have some initiative to be LLM friendly.

Getting back to the problem of people doing stuff that requires access to library materials: What sort of server-side cost multiplier do you suppose this ACM page has vs. just serving a few S3-compatible URLs that point to a few mirrors? (Or, got forbid, Anna's Archive / libgen, since ACM is open access, and those sites would be doing nothing wrong by mirroring ACM content.)

The EU banks are mandating that all their systems be backdoored by Google.

Trump has started forcing US companies to deplatform people in Europe that he does not like. This is not a legitimate anti-fraud step.

It means the spyware information flow and Google's ability to remotely modify the device have not been disabled.

This bank claims "security is very important to them". In another, the poster reports getting training spam every week.

Installing Lineage isn't that hard. If they're going to badly inconvenience some customers / force hardware purchases, why not start with the ones who are most likely to get their accounts drained?

Microsoft Amazon Google Apple to be clear.

I’m pro-AI. I use it all the time. I max out my $200 claude subscription and own a strix halo.

The data center buildout makes no sense.

I don’t want to work for a Trump lapdog, so MAGA++ recruiters can go pound sand.

I don’t think SV leadership realizes how f—-ked they are.

Without knowing which dots are which, this is useless.

Let’s say I want to run qwen 27B or flash next on a machine I own, off solar panels that are also sunk cost.

How did they price the marginal cost of a token? Depreciation on the box + reverse billing opportunity cost with the power company?