memoriuaysj

you are right, but you are misplacing the blame. it's not that you dont own your phone, it's that you dont own your bank account and the bank can dictate how you access it

the banks would care. less money spent on security or dealing with clients who had their money stolen

remember how all of HN and tech people were saying that DNT is a Micro$oft scam designed to break privacy because it was enabled by default without requiring user action?

to the point that Apache web server developers added a custom rule in the default httpd.conf to strip away incoming DNT headers !!!

https://arstechnica.com/information-technology/2012/09/apach...

we can use AI to rewrite everything in Rust

this way all the RAM that AI data centers scoop up will be used to lessen demand for RAM that those same datacenters created

net-zero RAM!

that's not enough, the tool definitions change, the agent harness changes, you need to pin a lot of stuff

sure, but you can have more memory controllers (I/O dies), like threadripper

bus wires. you can route only so many of them on a motherboard.

it's why GPUs have their memory chips in a circle around the GPU chip.

oracle free tier seems to be 1GB RAM.

why dont you run the VM on your machine?

when the agent pushes the PR, in a branch, you can switch to that branch locally on your machine and do whatever, review it, change it, and ask for extra modifications on top, squash it, rebase it

OK, i didn't know switching to Clicks requires so much time, I assumed using it is natural, I imagined is just a "raised 3D" touch keyboard.

but if it's like using a Dvorak and a Qwerty at the same time, it makes sense it should be the only device

the thing you need to trust is that pressing that combination shows the legit OS stuff, that it can't be intercepted (Secure Attention Key).

how the OS implements what is displayed is irrelevant

windows has all kinds of virtualizations today, it can literally run web views in separate (invisible) VMs for security purposes

what has gone horribly wrong is the native UIs. they are completely worthless, across all OSes - difficult to use, limited, and in general suck compared to HTML/CSS.

I've worked with all major GUI frameworks, from MFC to Qt, they all suck compared with React/Vue

you are arguing with math proofs here, the information is not gone, if it was a real video (as opposed to adversarily generated video)

some people type a lot, chatting, writing emails

not sure if that makes them a power user

why do you equate second device with digital detox?

what happened to use the best tool for the job? use a phone with a physical keyboard when chatting on WhatsApp and then switch to a regular phone for Instagram and browsing web. not saying everybody should do this but if chatting is your life...

some people even use a phone and a laptop at the same time, they are already a second device person, so they could be a three device person

yes, it can run IG and Tiktok, but you'll have a bad experience due to the screen aspect ratio, and small screen size

the power user base you mention is probably too small to sustain them long term

ah, sorry, you just want to control what apps other people can use

for video there is another constraint - time

hardware encoders (like the ones in GPUs) typically work realtime-ish, so they do minimal exploration of encoding space

you also have the one-pass/two-pass thing which is key for unlocking high quality compression

do you have so little agency that you can't stop yourself from installing TikTok?

what will stop you then from keeping your existing TikTok phone after buying this?

people still need their IG and TikTok

you can fight that, and lose (no market)

or accept second device status (for werk), optimize that use case, and be honest that it will not be the main device

it's gone in a single still frame

but across many consecutive frames, the information is spread out temporaly and can be recovered (partially)

the same principle of how you can get a high resolution image from a short video, by extracting the same patch from multiple frames

https://en.wikipedia.org/wiki/Video_super-resolution

this is exactly what "higher" compression levels do (among other things like bigger dictionary) - they try harder, more iterations, to find the optimum combination of available knobs for a particular chunk of data.

if you really care about structured output switch to XML. much better results, which is why all AI providers tend to use pseudo-xml in their system prompts and tool definitions

what if I told you the same issue is true for lossless plain compression like .zip files

the compressor (encoder) decides exactly how to pack the data, it's not deterministic, you can do a better job at it or a worse one

which is why we have "better" zlib implementations which compress more tightly

it's not gone, just more difficult to extract

video has certain temporal statistics which can allow you to fit the missing information

only true blurred white noise is impossible to recover

YCrCb can also be better for HDR or not - 4:2:2 vs 4:4:4

if you expand limited YCrCb to a large HDR range you'll get a "blurred" output.

Imaging converting 1 bit image (0 or 1, black or white pixel) to full range HDR RGB - it's still black and white

all the quoted Python code on the medium post has broken formatting

your comment above has the same broken formatting

does not inspire confidence if you can't spot such obvious breakage

You mispelled cassette

how do you feel about containers versus VMs?

their uploaded minds will enjoy the infinite AI slop though

land. compute what surface you need for 1 GW of solar