It’s interesting that only GET calls are metered. Is this a common thing to do?
I wonder if this is also being done to limit marketplace data being scraped with the API or limit how this data gets used by limiting low margin business models. Increasing the fees will have these effects.
I've been noticing this happen more lately. Intuit QuickBooks did this exact same thing for GET calls this year as well.
My theory is they see POST/PUT calls as adding value and are actions that customer need to perform and help the platform (adding products, inventory, etc). But GET calls they see as leaching off them and not core requirements.
The huge gap in this theory are things like Seller Fulfilled orders which you have to GET, but for Amazon they have always pushed hard for Amazon Fulfilled instead.
In the end I think this is all mostly to keep things/control in-house as much as possible. Some apps will shut down, others will pass the cost along the sellers. Sellers always lose.
It’s been some time since I have dealt with web scrapers but it takes less resources to run a regex than it does to parse the DOM (which may have syntactically incorrect parts anyway). This can add up when running many scraping requests in parallel. So depending on your goals using a regex can be much preferred.
Two thoughts here when it comes to poisoning unwanted LLM training data traffic
1) A coordinated effort among different sites will have a much greater chance of poisoning the data of a model so long as they can avoid any post scraping deduplication or filtering.
2) I wonder if copyright law can be used to amplify the cost of poisoning here. Perhaps if the poisoned content is something which has already been shown to be aggressively litigated against then the copyright owner will go after them when the model can be shown to contain that banned data. This may open up site owners to the legal risk of distributing this content though… not sure. A cooperative effort with a copyright holder may sidestep this risk but they would have to have the means and want to litigate.
As for 1, it would be great to have this as a plugin for WordPress etc. that anyone could simply install and enable. Pre-processing images to dynamically poison them on each request should be fun, and also protect against a deduplication defense. I'd certainly install that.
I think the post's argument is that we are on the way to something akin to China's social credit system (but not there yet).
> What we have aren't unified social credit systems…yet. They're fragmented behavioral scoring networks that don't directly communicate. Your Uber rating doesn't affect your mortgage rate, and your LinkedIn engagement doesn't determine your insurance premiums. But the infrastructure is being built to connect these systems. We're building the technical and cultural foundations that could eventually create comprehensive social credit systems. The question isn't whether we have Chinese-style social credit now (because we don't). The question is whether we're building toward it without acknowledging what we're creating.
Social Media getting big → larger perceived friend groups
Social Media getting big OR
screen time increasing OR
phones preferencing more limited forms of communication OR
… → more polarization
Maybe I missed it but it would have been helpful to know which confounds had been ruled out.
It could be due to a change in the PEW polling or the polling questions staying the same but the definition of terms shifting over time which caused the perceived increase in polarization. Or even the researchers’ definition of polarization which was stated as an increase in people stably identifying as either liberal or conservative. It is worth noting the article did say the PEW polling is supposed to be a stable source of data.
- The update now clears the shutdown log each boot.
> This led to the conclusion that a cleared shutdown.log could serve as a good heuristic for identifying suspicious devices.
> With iOS 26 Apple introduced a change—either an intentional design decision or an unforeseen bug—that causes the shutdown.log to be overwritten on every device reboot instead of appended with a new entry every time, preserving each as its own snapshot. This means that any user who updates to iOS 26 and subsequently restarts their device will inadvertently erase all evidence of older Pegasus and Predator detections that might have been present in their shutdown.log.
AI browsing the web is dumb AF if you think about it. Using an API through a REPL is so much better, we're doing all this work to basically work around jackass site operators who make everything require javascript and don't provide a documented user facing API.
The irony is that as the agentic boom really takes off, all these no-api, no accessibility sites are going to lose to small competitors who just offer a reliable agent interface, so people can use their service without having to use their service. Good riddance to the dinosaurs.
Obviously an API is better but realistically we aren't going to convince every web service to offer an API overnight and people want to be able to e.g. make reservations through chatgpt today.
Yep, the only way to convince companies to offer an API is to implement an agent that slowly but surely works around whatever trainwreck of a web experience they put in its way and then give them an option to make it smoother by offering an AI.
See: mobile websites. They sucked so badly that "desktop internet, not mobile internet" was a big selling point of the original iPhone. Then, once mobile had enough market share to "set the terms," we went back to having special mobile versions (or even mobile-first), but this time it didn't suck. Part of that was tech, but most of it was mobile acquiring a critical mass of marketshare, and the winner of the mobile wars won using an all-important temporary workaround stepping stone that solved the chicken-egg problem.
But maybe if you look from a first principles standpoint, do most human tasks decompose to some form of these same 4-6 tasks? (not talking about brainstorming, which is already well covered, or socializing, which is offline)
The only useful case I can think of is if you’re on a website with a big unstructured list or collection and you want to filter or reformat the data. For example, say you’re looking at a listing of houses for sale and you want to see only the ones that are painted blue, but the site doesn’t have that kind of structured data. Then AI could help by looking at the images and picking those out. Still, that’s probably not a very common situation, and you could do something similar with a bit of scripting and feeding that data into an AI manually. But for people who don’t know how to code, or are intimidated by it even when AI writes it for them, I guess it could be useful.
Oh and maybe one more thing to just give you the content that you're looking for like on all of these recipe sites with walls of text and images for SEO purposes where you just want the recipe. I guess that could be useful to just ask show me the recipe.
The demo looks like holding a robot's hand while they do something that would normally take me 15 seconds anyway. I have mostly found AI to be useful for search/research, not creating a middle-man between my friends and myself who has the "feature" of knowing what the star ratings on Google Maps imply.
I wonder if this is also being done to limit marketplace data being scraped with the API or limit how this data gets used by limiting low margin business models. Increasing the fees will have these effects.