This is a big underutilized dataset for mapping features, eg. parking restrictions on a given street or building entrances. With something like Gemini Flash, it's cheap to systematically extract map features from that imagery. Surprised that even Google doesn't extract much useful street view information into their maps.
Channels are a great first level, and iOS absolutely needs to implement an Android-tier version of this.
But channels continue to be abused, even on Android. When all deterministic controls fail...
Secondly, channels are set by the developer (or platform). In an ideal world, I want to define whatever channels I care about, and turn them on/off at will.
iOS really needs LLM-based notification filtering. This would take care of promotional notification spam overnight. It would even enable fine-grained user filtering like "notify when - someone is messaging me about plans for today."
The distinctions drawn here are particularly interesting in China.
Somewhere like Shanghai, you'll see ~70% of traffic in "bike" lanes are what appear to be electric mopeds.
But if you look closer, all of these mopeds technically have tiny attachment points for pedals. Government regulations allowed e-bikes to be driven unlicensed (but with a special green license plate, unlike the US!) and wherever bicycles are allowed. At the same time, the delivery industry and commuters wanted something stable, capable of carrying cargo/passengers. So the form factor adopted was that of mopeds, while vestigial pedal attachments were provided in order to pass as "e-bikes" under the regulatory criteria. Example. [0]
In practice, using pedals on these made for a clunky experience so they were not usually attached at all. The other main regulatory criterion was that these have to be limited to <= 25 km/h, unlike true mopeds/motorcycles. In practice, these speed limiters were also removed, setting up a cat-and-mouse game between police and riders.
The rule requiring the vestigial pedals was finally removed a few months ago, meaning that the ontology of "e-bikes" is pretty different in China now. [1] (Pedal-assist traditional bike frames also exist, but they share space with the larger mopeds in bike lanes and bike parking. True electric mopeds and motorcycles also exist, but they are effectively regulated out of existence in big cities.)
At the end of the day, top speeds are more determinant of whether different modes of transportation can coexist than pedals or form factor.
I wish there were a way to “archive” cards and passes in the Wallet app. I’d be much more likely to pass-ify my life if that were the case.
The Wallet app is just too important and used frequently in time sensitive actions to clutter with cards/passes that I use once every few months. That is, when I’m about to tap to pay, I don’t want to infrequently used cards to clutter my payment experience. Likewise, when I’m about to board a flight, I don’t want random loyalty cards to clutter the interface.
At the same time, I would really like to keep these occasional cards and passes in Wallet, just not on the main screen. It definitely beats hanging onto these physically, especially because they are in fact infrequently used so I would never carry them around.
It should be a similar distinction to Apple’s Home Screen vs App Library for long-term archival.
I feel like street-view data is surprisingly underused for geospatial intelligence.
With current-gen multimodal LLMs, you could very easily query and plot things like "broken windows," "houses with front-yard fences," "double-parked cars," "faded lane markers," etc. that are difficult to generally derive from other sources.
For any reasonably-sized area, I'd guess the largest bottleneck is actually the Maps API cost vs the LLM inference. And ideally we'd have better GIS products for doing this sort of analysis smoothly.
Moving to a subscription fee model is dangerous for a service that's reached saturation point. They can no longer project infinite growth in value extracted from eyeballs going into the future, and are stuck with whatever they choose to charge (along with some limited increases in fees. Building a better service cannot be fully leveraged into higher fees, since there's only so much YoY increase in costs users will stomach.)
On the other hand, subscription models are fine for new entrants, even with VC hyper-growth expectations - there are still billions of users to capture even if per-user revenue is fixed! Then the game becomes delivering as much value to these customers as possible to attract more paying users.
Perhaps this is another malincentive that comes about from monopolies.
There is in fact significant doubt among climatologists as to the veracity of this record.
While the 1913 hot spell was certainly significant, all recorded weather information at the time from surrounding locations, in combination with the dynamics of local microclimates, imply that the 134 degree reading was "essentially not possible from a meteorological perspective." [1]
Observer error stands to be the most plausible explanation, implying that the the recent 130 degree reading in Death Valley may actually become the location's, and the world's, hottest verified temperature.[2]
The usual problem with this pedagogy is that kids get confused when presented with improper fractions (eg. 11/8), and so on. I guess this is probably still a good intuition to start with, but how did you tackle those extensions later on?
I'm a little concerned about the new Safari interface [0], which very smoothly integrates the tab/address bar with page content. For example, the background color of the page flows behind the open tabs such that it looks like one unified interface, rather than browserchrome || pagecontent.
This is all great to experience as web apps increasingly take over the functions of native apps. It does help them feel more like first-class citizens, rather than plain documents pulled up through a program.
But it easily brings up new potential abuses by phishing sites, spammy notifications, and other bad actors. The new design seems to start breaking down the browser UI's Line of Death [1], at least in perception.
Naspers' market cap is 98B, yet they own 31% of Tencent which is worth 775B - am I missing something? I'm aware of conglomerate discounts but I suspect there's something else going on here.
As others have mentioned, it does depend on what exactly you're defending against.
Preemptively opening the link as the sender will send a request to TikTok, but they're not really gaining any useful data there since you just watched the video, hit share (this is what they know so far), and now you opened the link that you had generated. So their database only learned that you shared a video with yourself, which you immediately opened.
The more valuable data is when various intended recipients open the link, allowing TikTok to associate you with them to serve more targeted videos based on implicit social graph, etc.
Moreover, opening the link yourself to get the "canonical url" protects yourself if you're sharing the link broadly since others can't obtain your name [and potentially more?] from the shortlink.
Now, if you're the recipient, there's not much you can do to avoid the tracking link, besides opening it up in as much of an anonymous environment as possible. But interestingly enough, I find the privacy threat greater to the sender. The sender has a TikTok account to aggregate data quite straightforwardly, unlike the recipient. The sender is also being associated with a number of recipients, vs. the recipient with only one sender, and again only through cookies, IP, or something of that sort.
A particularly bad instance of link tracking I've found is in TikTok's link sharing feature.
If you share a link from the TikTok app, it gives you a vm.tiktok.com/[xyz] link to send/post elsewhere. It gives you no indication that this isn't a generic link to the post, nor does it give you an option to expose the generic link to the post.
Instead, when you share that link and someone clicks on it and does not have the app, it opens with a header saying "[First Last] is on TikTok." On the other hand, once you do click on that link (if and only if you don't have the app installed), you get redirected to the static link to the video and finally obtain it.
This is an anti-pattern that enables further tracking and potentially unknowingly exposes user data when links are shared publicly. And there's no indication to the user that this is happening, since the link is structured as if it does not contain any tracking. Ie a tool like this wouldn't be able to "strip out" the tracking since it isn't tacked on in any way, but embedded as the generated link itself.
Related, another paper using CLIP with StyleGAN for text-based semantic image manipulation from just a few days prior ("Paint by Word", where the user can select the area to be transformed): https://arxiv.org/pdf/2103.10951.pdf
I don't think we could come up with a standard x axis bounds for such a graph, since n=1000, or 1,000,000 may not be zoomed out enough to showcase the behavior approaching infinity.
