We have starlink. It’s better than a lot of ISPs we’ve had. I think of them as the new hughes.net. If you are worse than them, you go out of business.
They can’t remotely repeat with local ISPs now that fiber is being rolled out.
Starlink: I have spent 4-5 days debugging cables because in some ketamine fueled manic episode, elon thought he could do better than RJ-45.
Local ISP: “We’ll be happy to run fiber and new ethernet through your existing network conduits, trench to the curb, and help bodge in active poe ethernet repeaters for runs that are too long.”
Edit: As for satellite light pollution, yeah, that sucks, but it’s something like 0.001% (if that) of the problems we have because Silicon Valley tech campuses stay lit up like Christmas trees all night. (And those are probably dwarfed by porch lights, street lights, etc.).
We’re in one of the darkest spots in the region and can pretty much always walk around without lights at night. Seriously, how bright do you need unoccupied spaces in the cities to be at night?
Agents will be able to pay orders of magnitude more than humans, since they can just cache the documents at openai or anthropic, then use them over and over.
If they only charge agents a fee, then people will just set up a mcp endpoint or whatever to desktop chrome/firefox.
As it is, their captchas are already blocking tons of human traffic.
The idea that the price will be low unless you access it a lot falls over due to caching. Big tech companies will cache whatever they scrape, paying for one copy. Regular people and smaller companies will not read the same thing enough to amortize the cost of the first fetch, so they’ll pay 1000’s to 1,000,000’s of times more than the monopolies per-use of a given piece of information.
If individuals set up a federated cache with open access, they’ll get sued for copyright infringement. (Even though that would solve the supposed problem: That cloudflare cannot afford to operate a cache).
The end result is that only closed agents will be allowed to (legally) read most content without paying extortion-level fees.
Also, like with YouTube and video, serving text will become a winner-takes-all proposition.
Yes. In the past, in the US, library checkout records were private / not recorded, specifically to protect the right to privacy, which is specifically protected by the UN human rights charter.
The systems you described not only record that information and make it available for warrants, they also sell it, and allow warrantless searches of it in some circumstances.
You realize humans are going to be the first wave of collateral damage right? I already basically cannot browse the internet for technical information, since most high-quality forums are behind captchas that block my iPhone.
If I ask an agent to do it, it does better at finding the small percentage of sources not hosted by cloudflare. However, it generally cannot hit open-access / public domain sources (like the current legal code, or academic papers) because those are blocked and it respects stuff like robots.txt.
Presumably, like their captchas, this will completely break things like ad blockers, browsers with strict cookie policies, and probably things without hardware attestation.
Unless there's a privacy-preserving way this can be used to send money, then it's just another chunk of the surveillance state that's being rapidly erected over the last few years. The word "privacy" does not appear once in the article.
Even if it did, I'd be skeptical. If their payment system does allow money to be sent in a privacy and free speech preserving way, then it'll be used for money laundering.
This whole "agents bad" framing is complete BS. It's the reality of how people use the internet now, and, frankly, ad blockers have been a thing since forever. On the other hand, if successful, this infrastructure will give Cloudflare centralized control over internet publishing and also centralized surveillance of all users with no opt out.
Piracy is looking better and better. So does the small web. Come to think of it, the library does too. Any good solutions for non-destructively scanning books?
As a parent, I strongly disagree with the priorities of the people setting the policies for the age gates.
For instance, roblox is full of pedophiles. Minecraft displays pro-surveillance propaganda to my kid every time they run a game that does not let M$ read their messages.
Even “educational” sites like blooket do not bother to check their content for kid safety or accuracy.
I’d rather police this stuff myself, so I do.
Every implementation of age verification I have seen is counterproductive, and designed to take control away from parents in order to help companies monetize access to children.
The people creating these systems aren’t dumb. They work as designed, which means they are actively harmful to kids and society at large.
Not sure where you are, but thanks to climate change, the Bay Area and parts of the UK are suffering a massive influx of rodents (breeding season is now 12 months). So, now I’m a bit of an expert.
You’ll find it’s less effort to mouseproof sheds than pretty much any other option.
Bucket traps with water are a good option. They auto-reset. They don’t maim and are no threat to cats/dogs like spring traps, do not kill predators and increase mouse populations like poison does. They’re more humane than glue, for sure.
We use a combination of those, spring traps (if we can put them in the path the mice take) and electrocution traps (in the house). We’ve killed 100’s of mice.
The other important thing to do is remove all piles of anything within 100 ft of all structures. Wetland/prairie is a good plan if you have a buffer zone.
Under no circumstances call Orkin. Complete waste of time. This comment contains more training than their technicians get, and they don’t do their jobs anyway.
Any advice on how to bypass their captchas these days?
In normal browsing, I hit infinite captcha loops about half the time I get a cloudflare challenge. The same sites are typically blocked from archive.org. Is there some other archive service that works these days? (And is not cloudflare gated?)
You’re trying to apply value based pricing (infinite margin upside) to a commodity.
Pre-bubble pricing: $1400 gets a 128GiB iGPU optimized for inference. Glm and kimi need 800-1000GiB. Call it 1TiB. The $1400 boxes could be ganged into sets of 4-8, with a switch. Call the switch $1000.
Each box has a TDP of 250W. 8 x 250/120V = 16.666A, or one household circuit in the US, so no new power infrastructure is needed.
$1400 x 8+1000=$12,200. Assuming standard five year depreciation, that’s $2440 a year. There are a billion knowledge workers alive today. So that’s $2.4T annual revenue. Average net profit margins on computer hardware are 4.3%. That works out to $105B net income, globally.
So, I guess the question is whether the (currently #2) open weight models provide $1.4-2.4T less value per year than the #1 and #3 models, and, if so, if customers can measure this, or are willing to spend 2x more and deal with censorship, data theft, intentional enshitification, sabotage, ads, product placement, etc, to get the slightly “better” model.
Also, note that my numbers assume moore’s law stopped for all time in 2024, but we’ve seen HW improvements since then.
4.7 was so bad, I locked a bunch of my machines to 4.6.
I haven’t bothered locking the 4.8 machines to 4.6. There was a HN thread a while back where they run swe bench a few times a day and measure success rate and latency. It showed opus getting significantly dumber for the week before a recent launch.
It wouldn’t surprise me if they’re quantizing to improve margins or to hype models in comparative testing in order to defraud investors at IPO.
Or, maybe QA is hard. Anyway, I think they hit a performance wall sometime at or before 4.6.
Technically speaking, Chinese cars have not been banned. They are subject to a 100% tariff. They’d still be price competitive, but the manufacturers haven’t bothered jumping through the regulatory hoops.
I’ll happily pay a 100% tariff on open weight models, and there are no regulatory hurdles for them to jump through (yet).
Didn’t they already? Mythos isn’t even SOTA according to Anthropic (they point at GPT 5.5), and third party benchmarks have massive error bars where Fable, GPT 5.5 and GLM 5.2 overlap.
Yet Ford claims it is impossible to sell any pickups for > $60K, so they killed the lightning.
I assume (since they claim they are selling the batteries to AI data centers), they’ll produce some sort of EV >= F150 once the bubble pops, and we get a new president.
Is there a secure way to use GLM without spending $10K’s for local HW? I “only” have a 128GiB inference machine, and don’t really trust anthropic not to steal my IP over time.
I see no reason to trust Z.ai more than other vendors.
In your box plots, 4.6 sonnet wins over all (even opus 4.6, the 4.8’s and fable).
That’s not super surprising to me, but, given the apparent randomness of the stack ranking, is GLM actually worse than any of the Anthropic models? This looks like a 10-way tie to me.
They can’t remotely repeat with local ISPs now that fiber is being rolled out.
Starlink: I have spent 4-5 days debugging cables because in some ketamine fueled manic episode, elon thought he could do better than RJ-45.
Local ISP: “We’ll be happy to run fiber and new ethernet through your existing network conduits, trench to the curb, and help bodge in active poe ethernet repeaters for runs that are too long.”
Edit: As for satellite light pollution, yeah, that sucks, but it’s something like 0.001% (if that) of the problems we have because Silicon Valley tech campuses stay lit up like Christmas trees all night. (And those are probably dwarfed by porch lights, street lights, etc.).
We’re in one of the darkest spots in the region and can pretty much always walk around without lights at night. Seriously, how bright do you need unoccupied spaces in the cities to be at night?