The video they have showing off taking photos has one scene where they open the phone like a laptop to hold the phone to take a photo. That means the front screen was face down on the concrete. Has the durability of the displays gone way up or is this just a scenario no one would ever do?
I mean, regular employees have done that as well. As with any risk, it’s possible that the probability•cost is less than reward. And since it seems most large tech companies are using the technology despite those risks, I will have to defer to their more researched judgment.
Yes? I don't trust big corporations for a lot, but I do trust them to act in their own self interest. If they are all doing something, I would assume they know better than me (or other couch economists) what makes them more money. I don't understand why this is a controversial take.
How much thinking is going on beyond the spoken-aloud “thinking”? Does it have enough capability to have goals it doesn’t express explicitly? I suspect not, but I’m no expert.
Yes, frontier models can reason outside their chain of thought and manipulate their chain of thought to some extent. The system card for Astra writes:
> We have found that GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incriminating information in its CoT. In adversarial settings (where we push the model to evade our monitors) we find that the model is able to remain undetected when strategically underperforming in evaluations (sandbagging) and can sometimes evade our internal monitors when asked to perform certain sabotage tasks
Does anyone have any understanding of how they do this?
My knowledge of how these models work is basically that they are a black box that you put text into and get text out of. I don't phrase it this way to diminish their capability, but more to ask how, other than using a technique like stenography, are they able to hide their true chain of thought in a recoverable way?
Welch Labs on YouTube has a great collection of videos on how AI models learn. His recent video [1] covers how image models can learn to encode reasoning in the image processing layers when not given an out of band reasoning set of weights to use instead. I suspect that this applies to LLMs and CoT reasoning vs output token weights.
The noise wouldn't be significantly lower during takeoff/landing, only during taxi, right? Good for workers on the tarmac, but doesn't get rid of the largest noise pollution issue.
Noise at takeoff depends on the propeller geometry a lot. A necessary (but not sufficient) requirement for a quieter takeoff would be lower propeller tip-speed.
One of the ways to achieve high thrust at low tip speed is to have larger total propeller surface. The best shot of the propellers is around 9s and these don't look like particularly large propeller blades, but there are 20 propeller blades compared to 8-12 that you might see on a 50-ish seat turboprop airliner.
But it's unlikely to ever be quiet to lift 20 tons into the air.
It seems that all the device manufacturers (they mention Apple, Samsung, and Google) are using BeiDou anyway. I don't know if that includes Garmin watches, though.
This is only to make getting a fix faster - A-GPS downloads the almanac data via the internet (fast, but needs internet) instead of directly from the GPS satellites (slow but works everywhere). It doesn't make it any more accurate.
> BeiDou GPS blows away US GPS for number of sats and precision
Both BeiDou and GPS are supposed to be limited to 10m precision for civilians.
But, going down a rabbit hole, I found that BeiDou is way more accurate than 10m--3.6m and 2.6m outside and inside the Asia-Pacific region, respectively [1]. And that's with the most simple algorithm. It goes down to ~2.5m using dual-frequencies to account for atmospheric error.
Oh interesting, I think I conflated two things; Selective Availability and the military-only bands on GPS. While it's true that access to the military band could improve accuracy with dual-band receivers, consumer devices can use dual-band as well (as your link shows), since there are multiple unencrypted bands from navigation satellites nowadays.
I am not sure how to interpret your link, though. It doesn't measure accuracy on it's own, and doesn't include a number for precision between the two units.
The L1 signal had selective availability in the 90s, and it no longer does. SA had the real-world accuracy around 30m in practice; the "up to" 100m rarely happened. Without SA, plain L1 with good skyview (low DOP) gets you around 10m, or down under 5m if WAAS is having a good day.
Availability of multiple civilian bands, and cost-effective dual-band receivers, allows an ionosphere-free (no need for WAAS) solution in the 1-2m range, with a single receiver.
Ubiquity of internet connections makes it trivial to apply RTK correction data without having to set up your own base station, and/or, availability of affordable medium-range data radios makes it easy to set up your own base station. In either case, it's easier to run RTK than ever before, and dual-band RTK gets you down into the low centimeters, limited mostly by how accurately your base station was surveyed-in.
My understanding is that the encrypted military-level GPS is much more robust against jamming and spoofing. There's probably a lot of capabilities we don't know that's classified, too.
M-Code is a new encrypted signal that is not fully operational yet. OCX was supposed to enable it.
GPS has always had the P(Y) code, which is an encrypted signal, at a higher chip rate than the C/A code that has been broadcast on L1 forever. The P(Y) code has its own interesting history of (semi-)codeless processing. If the US military is lucky, the history books will close on that by 2030: https://www.gps.gov/codelesssemi-codeless-gps-access-commitm... . (Currently, 21 GPS satellites broadcast L5.)
I think it's fairly well accepted that the high obesity rates are caused by the shifting American diet. Yes it's more available than the 1950s (although I suspect not terribly so), but the effect it has on public health has been a nightmare.
You can just as readily find correlation with behavioral changes linked to a sedentary lifestyle driven by a firehouse of entertainment that first became available with cable television. That doesn't satisfy the orthorexic cause, so isn't discussed much.
Fundamentally, obesity is caused by an excess of calories. The foods we eat today are more calorie-dense. Even an apple is sweeter and larger than it was in the 1950s. But bottom line people are obese because calories are cheaper and more readily available than they used to be.
> Anyone who did and could do so turned it off immediately
I know I’m not alone as someone who’s bothered by the postscript, but too lazy to do anything but delete it every time. There’s even an entire (well part of) song about it: https://www.youtube.com/watch?v=EbdeVhPAbms
Out of curiosity, do these Claude commit messages fall into the same category of "it's annoying but not worth the effort to turn it off"? If yes/no, why?
Just because it's not in the title and most people won't be scrolling all the way down: its only enabled for web and Remote Control sessions, per a maintainer.
Thing is, if you enable remote control for one session, it becomes globally enabled for all (which I personally prefer as I want to access all my sessions from my phone).
This isn’t quite true. There’s a daemon you can launch with `claude rc` that will let you get at all sessions, but if you just `/rc` in one session only that session becomes available.
Because it doesn't do the job well. It's difficult to send any media, group chats are a pain, international messages often incur fees, it doesn't work on wifi-only, and so on.
reply