Agreed. It looks like over-aggressive adaptive noise filtering, a smoothing filter and some flavor of unsharp masking. You're correct that this is targeted at making video content compress better which can cut streaming bandwidth costs for YT. Noise reduction targets high-frequency details, which can look similar to skin smoothing filters.
The people fixated on "...but it made eyes bigger" are missing the point. YouTube has zero motivation to automatically apply "photo flattery filters" to all videos. Even if a "flattery filter" looked better on one type of face, it would look worse on another type of face. Plus applying ANY kind of filter to a million videos an hour costs serious money.
I'm not saying YouTube is an angel. They absolutely deploy dark patterns and user manipulation at massive scale - but they always do it to make money. Automatically applying "flattery filters" to videos wouldn't significantly improve views, advertising revenue or cut costs. Improving compression would do all three. Less bandwidth reduces costs, smaller files means faster start times as viewers jump quickly from short to short and that increases revenue because more different shorts per viewer/minute = more ad avails to sell.
I agree I don't really think there's anything here besides compression algos being tested. At the very least, I'd need to see far far more evidence of filters being applied than what's been shared in the thread. But having worked at social media in the past I must correct you on one thing
>Automatically applying "flattery filters" to videos wouldn't significantly improve views, advertising revenue or cut costs.
You can't know this. Almost everything at YouTube is probably A/B tested heavily and many times you get very surprising results. Applying a filter could very well increase views and time spent on app enough to justify the cost.
The most realistic acquirers were Paramount/Skydance or Netflix. Paramount/Skydance is a relatively new-ish entity with David Ellison (Larry's son) as CEO. The general sense in Hollywood is Paramount/Skydance will do little high-brow, art house or awards-fodder films but they will at least distribute films primarily to theaters (they promised to release at least 14 Warner films per year to theaters if their bid was accepted).
Netflix is mostly uninterested in theatrical distribution so the main practical impact of this most of us see day to day may be less theatrical release movies and probably fewer higher budget films being made at all.
Caveats include that the deal has to actually get regulatory approval in the U.S. and EU and survive potential (inevitable?) shareholder lawsuits. Netflix's offer reportedly involved less cash and more debt. Paramount/Skydance argued regulatory approval and the heavy debt made Netflix's offer less attractive than their own despite Netflix's higher top-line price.
> it is possible for a savvy user to partially control the newer hardware in limited ways by hacking open the wall button and soldering a connection to a third-party relay module that interfaces with your home control system. But your home control system won't know if the garage door is open, closed or in-between or if/when it was remotely activated from an arriving or leaving vehicle. It also won't know if the door stopped due to the electric eye being blocked. Knowing the full system state, activation history and sensor data is much better, enabling all kinds of flexible automations like auto opening as you enter your driveway.
You could, in theory, start adding more third-party sensors around your garage door opener like a tilt/angle sensor on the door, a voltage detector on the motor wires, etc but it's just more and more "science project" to regain access to information that should be yours to start with. It's not like Chamberlain is giving away these garage door openers for free - just buy a non-Chamberlain brand (note: they now own MyQ, LiftMaster, Merlin, Arrow and several others so shop carefully). If you have a 2.0 (or earlier) Chamberlain from before ~2022 get a RATGDO to open it back up 100%. If you're stuck with a post-2022 encrypted 3.0 Chamberlain, then your only options are science project hacking to regain some limited control or just replacing it. I'm fortunate to have 2.0 Chamberlains and I only bought them because they HAD an open API that worked with open source Home Assistant - then Chamberlain sprang their carefully laid trap. I don't believe it was "accidental" that they heavily marketed compatibility with open source while acquiring several competing brands, then discontinued API access at the moment they released new hardware with unnecessary encryption on wires that are internal to the user's home.
Yeah, having a system that knows the door's state would be ideal, but TBH we have cameras that can see the state already, so just checking that before hitting the button would do the trick for me. I recall seeing a smart device that can be used to push buttons on coffee makers or garage doors, or other devices that require physical button presses.
Yes, there are several cheap Home Assistant-compatible devices to do dry contact closure and/or a 5v pulse. The Shelly 1 is a popular turn key Wifi-based option but there are wired and Zigbee options as well.
Many come with (or can be OTA flashed with) Tasmota or ESP Home open source firmware, ensuring you'll continue to control them forever locally and/or via your cloud API of choice - which I highly recommend doing. I built a new house with 75 ESP 8266-based in-wall WiFi dimmer switches ($19 ea) plus dozens of other wall plugs, power strips, sensors, etc - all running Tasmota. It's great because each Tasmota device powers up and works as expected as a 'dumb' device. The dimmers always dim and the wall plugs/power strips have physical on/off buttons that always work.
But they'll also connect with each other via Wifi in ad-hoc device groups (if I enable that option) and will follow whatever rules I set in their on-device config - like "you are Device #3 in Group #2 and you follow-the-leader of Device #5". And they do this even if there's no Wifi router active. If there IS a Wifi router available, then they'll log on to the local-only subnet I've limited them to and work under local and/or cloud Home Assistant control with infinite automation possibilities. So it's an open, flexible, secure system that can be maximally "Smart" but has layered fallbacks which guarantee as long as there's 110v power - the lights, plugs and other devices always fucking work.
IMHO, open source with layered dumb --> local --> any-cloud-API-you-like is the only sane home automation architecture. There's no way I'm ever permanently installing some for-profit company's opaque, remotely updatable system into my home's walls. Even if they don't turn evil like Chamberlain did, it would be crazy to leave the basic functionality of my house's lights, door locks, HVAC, sensors, etc at the mercy of some vendor bug, broadband outage or regional S3 'mis-configuration'.
Chamberlain is one of the worst anti-user companies because they used a trojan horse deception while they were buying up other garage door opener companies to eliminate competition (they now own MyQ, LiftMaster, Merlin, CPSG, Arrow and several others). Their products used to have an open API and worked well with open source home control systems like Home Assistant. HA support was specifically why I chose to install their products in two homes (Liftmaster brand). Then around 2022 they pulled a Darth Vader and "changed the deal" by shutting down API access for third-parties - except for companies willing to pay them - AND releasing a new 3.0 version of their control hardware that's encrypted.
While the prior versions of their control hardware weren't documented, it was still possible to reverse engineer the serial protocol. But the new hardware even encrypts communications between the wall button and lift motor for no legitimate security reason other than "securing" Chamberlain's revenue stream. Fortunately my units are the older 2.0 version hardware so third-party add-ons like RATGDO can tap into the system and open it back up. While adding that hardware is an extra hassle I didn't expect, newer customers don't even have that option. I used to have a Home Assistant failsafe automation that closed the garage door if it was accidentally left open and we weren't home. That automation worked great for years and then just stopped working when Chamberlain disabled their open API.
Even if Chamberlain provided a similar capability in the set of minimal "free features" in their walled garden of paid upgrade app features (they don't), I wouldn't use it because they lace their apps and panels with pop-ups ads. And there's ZERO fucking chance I'm giving some company's black box device access to my family's real-time phone/vehicle locations or internal home motion sensor data to enable the most useful deep automations. It's bad enough their untrusted device is still in my home. There's no way I'm giving it access to my network, much less the Internet.
(To be clear, it is possible for a savvy user to partially control the newer hardware in limited ways by hacking open the wall button and soldering a connection to a third-party relay module that interfaces with your home control system. But your home control system won't know if the garage door is open, closed or in-between or if/when it was remotely activated from an arriving or leaving vehicle. It also won't know if the door stopped due to the electric eye being blocked. Knowing the full system state, activation history and sensor data is much better, enabling all kinds of flexible automations like auto opening as you enter your driveway. Never buy a garage door opener from a Chamberlain-owned brand).
Been planning to build my daughter a gaming PC for Xmas. Feel lucky I found a pre-built with 32GB RAM in a warehouse store on black Friday that hadn't been marked up yet. First pre-built desktop I've bought in decades and it was actually cheaper than I could build myself due to the crazy RAM prices.
Unfortunately, TFA is yet another article promising an answer to a "Why" question it never delivers. It documents the issue with many examples and thoroughly explores the variations yet never answers why current LLMs write as they do. It mentions some ideas in passing which might be causes but provides no justification or expert input.
Meanwhile, I'm pretty sure there are LLM researchers exploring this problem not only because it's annoying but because it's also interesting. If anyone knows of papers on this topic, I'd be interested...
I don't know papers but I figure the main reason why it's odd is humans have a picture or idea in their heads and want the reader get the picture or get the idea. LLMs don't and just string words together so it starts reading normally and then kind of wanders off.
Like for me the first paragraph reads normally up to "whisper woven from the algorithm" but then "a construct of code. A.I.-generated writing..." is just kind of muddying the picture with excess waffle.
It would be deeply ironic if this data center (or similar ones using creative accounting), are among those featured in the TV commercials Meta has been running in expensive national prime time slots in recent weeks.
I've seen at least two different commercials each focused entirely on the personal story of a relatable, folksy person living in a small town in a fly-over U.S. state, talking about how the town was declining and times were hard - then Meta built a new data center nearby and this person along with many others got jobs there and now things are great. They are very well-produced with cinematic shots of rustic small-town main streets, dusty pickup trucks in rural settings and local high school football games. Aside from the obvious brand-washing, it would be extra on-brand if it turns out Meta doesn't even own the data center but still tries to take credit for it.
These facilities will sometimes employ as many as 100 people - so a state that can attract three such data centres creates almost as many new jobs as an entire wal-mart store. Truly, a transformative number of jobs.
Yep. Massena, NY totally got hundreds of jobs out of the shitty Bitcoin miners that came to town like Coinmint and whoever the hell "North Country Colocation Services" is. We were saved! Everybody is rich and healthy and happy!
... wait, no, we're not. We're still an absolute shithole at the top of New York, now with a bunch of sea cans sitting in front of abandoned industry we lost decades prior, humming away doing nothing for any of us.
Most will be specialist jobs too, with people moving there to fill them and/or working on a rotating basis, requiring traveling. Truly magnificent for the community, and the environment--employees can bring bottled water with them from out of state to solve that issue!
It turns out the one in this ad is in Altoona, Iowa. The ad focuses on how it revitalized the community by providing jobs, kind of glossing over how that might be reflected in the massive facility's ~30 car parking lot.
And incidentally, that data center currently shows no open positions on Meta's career website, although third-party sites still have some dated listing for advanced IT positions that were probably filled by non-locals.
30 sounded low to me, but looking at the sprawling Altoona facility in Google Maps https://maps.app.goo.gl/KGLEpJRFiwVKYob89 satellite photos show 52 parking spaces in use across 11 buildings.
Lots of construction workers in the areas where they're putting up new buildings, though.
That data center started operation over a decade ago. There's no reason to assume it should have any vacancies currently, nor is there any reason to assume its positions were largely filled by non-locals. And even if a few external folks relocated in some cases, isn't that economically beneficial to the area regardless?
Also given the 24/7/365 nature of data center operation, the number of employees will be larger than the number of parking spots.
Meta says over 400 people work on-site at the Altoona facility [0], but most of those are clearly working for a variety of smaller contractors given that the initial tax terms anticipated a few dozen direct employees [edit: wrong state] and no datacenter companies show up in the county's 50 largest employers [1].
What? Me exaggerate? Never! :-) I did actually spend a minute looking for the commercial on YouTube but YT search sucks, so thanks for finding it. Setting aside the fact it's crassly manipulative corporate propaganda, as someone with a lot of film and video production experience, the production team that made it did very nice work. That's the only reason I actually ever even saw it. I normally skip through all commercials on the DVR, but I happened land on a couple of really nicely shot frames as I was skipping and rewound to see the spot. I thought, "Wow, blatant bullshit but great work!" :-)
"If you add two pounds of sugar to literally one ton of concrete it will ruin the concrete and make it unable to set properly which is good to know if you wanna resist something being built..."
Quick tip I discovered when traveling with my teenage daughter: a lot of hotel sites are now unclear on whether a booking is for a room with one or two beds. I found that listing "occupants" as 3 would usually force such sites to sort for rooms with two beds (even though there would only be two of us). Assuming there's no breakfast included, the price is usually the same for 2 or 3.
What country is this? I've never seen a hotel site that didn't sell rooms as either 2 Queen or 1 King. If I didn't know it was a king bed I wouldn't book it. Does that now make me a spoiled first world rich person?
Generally, what I’ve seen is on travel sites like Priceline, sometimes they list a room as like “standard room” and they don’t specify and (in the fine print) explicitly do not guarantee how many beds - with some cheaper rates. Basically trying to discourage people from booking them. The thinking being if you don’t wanna end up in 1 King bed with your bro, you’ll pay the extra $13 for the explicitly 2-bed room, which is always listed as well.
Which is odd, since I feel like I always end up paying $13 extra for one king bed with my girlfriend to make sure we're not sleeping in a queen next to an empty one
This particular trip was in Europe but I also encountered it on a different trip to Las Vegas. It occurred on some hotel sites but quite a few hotel aggregator sites.
So like, in the United States, if you book directly via Marriott, the number of beds isn't guaranteed unless you have some reasonable status at the hotel.
If you book a queen room isn't it always 2 beds? A king room is usually 1 bed. Is there some option where it's just totally random what room you get? I don't have any Marriott status but going on their site I can clearly see a choice of rooms, and each one says what amenities it has.
The point is that you aren't guaranteed the room type you select: you'll show up at the Marriott and they'll give you a room with one king instead of the room with two queens you thought you booked; your selection is just a preference, not a lock.
^ you have to have Platinum status to get the type "guaranteed"... and, even then, they might just give you $25-100 and say "sorry" (it is more of an SLA with a penalty; though, this is also true of the reservation itself... FWIW, getting a random room type is MUCH more common than getting bumped).
Too true. If you have a normal reservation (vs some kind of pre-paid room), they oversell their capacity based on a statistical model. This means if everyone with a reservation shows up, some of the last to check in won't have a room (which tends to occur during statistically unusual events like a large trade show, where every reservation is more likely to show up). The hotel will basically just say "sorry" and suggest another hotel which may have rooms. I think they try to recommend hotels that are nearby and similar but sometimes that either doesn't exist or is also sold-out.
One of the benefits of having platinum status at Marriott is they'll actually guarantee your room but platinum status isn't easy to get. I used to frequently drive into the bay area late at night to avoid traffic, arriving at the hotel at midnight or later. Being platinum at the time, I'd sometimes get a call on my mobile from the manager after 9p to find out if I was showing up for sure and I'd tell them "Yep, I'm driving now." That's how I knew they were sold out and bouncing people with reservations. They didn't want to keep holding my room if I wasn't going to show up (being platinum there was no penalty for no-showing on a reservation).
A tip: if you have a non-guaranteed reservation and you think the hotel is likely to be full but you're arriving later in the evening - call the actual hotel front desk before 5p (not the reservation center), ask to speak to the manager on duty and ask them to check you into your room over the phone. Ask for the actual room # and then have them actually print out your card key and put it in an envelope under your name at the front desk. Once your reservation has been checked in, a room assigned and a key printed, it can't be given away by the late shift.
Heeey clever. I really struggled with this while travelling with my brother in Japan. None of the aggregation sites filtered on number of beds even though they had that data in the listing.
I’ve been to Los Angeles recently, and they wanted to give us a single bed room for 3 of us, and they told us that “some” wants the one bed option for 3 adults for whatever reason.
The people fixated on "...but it made eyes bigger" are missing the point. YouTube has zero motivation to automatically apply "photo flattery filters" to all videos. Even if a "flattery filter" looked better on one type of face, it would look worse on another type of face. Plus applying ANY kind of filter to a million videos an hour costs serious money.
I'm not saying YouTube is an angel. They absolutely deploy dark patterns and user manipulation at massive scale - but they always do it to make money. Automatically applying "flattery filters" to videos wouldn't significantly improve views, advertising revenue or cut costs. Improving compression would do all three. Less bandwidth reduces costs, smaller files means faster start times as viewers jump quickly from short to short and that increases revenue because more different shorts per viewer/minute = more ad avails to sell.
reply