A number of the PM 2.5 studies I've seen do exactly this. They gather air quality data and run the figures based on preexisting models of health impacts to entire populations or even the entire planet and surprise out comes insane figures.

I just want to know how they came to radically different figures in a later revision of the same work. This is well outside the range of the CI in the version published in nature.

Prior version of this work here:
https://www.nber.org/system/fi...

It states: "We project that climate-driven increases in future smoke PM2.5 could result in 27,800 excess deaths per year by 2050 under a high warming scenario, a 76% increase relative to estimated 2011-2020 averages. "

From abstract published in nature:

"We project that smoke PM2.5 could result in 71,420 excess deaths (95% CI: 34,930 - 98,430) per year by 2050 under a high warming scenario (SSP3-7.0) - a 73% increase relative to estimated 2011-2020 average annual excess deaths from smoke."

This is why you should say please and thank you around LLMs and avoid unnecessary chatter.

DeepSeek the model (as opposed to DeepSeek the online service which is pretty bad) is in my experience no worse in terms of censorship/refusals than the other models.

Higher frequencies are preferentially scattered. Most LED street lamps operate at higher frequencies than the HPS they replaced.

They make shit up because they have no meta-cognition and don't know any better.

This study applies model of lighting conditions throughout the country to a circadian health model and draws conclusions about health impacts of each of the three scenarios from model evaluation alone.

I am personally skeptical of these types of extrapolations from models generally. Numbers being bandied about in this study seem unreasonably large. I would be far more interested in learning about studies that actually managed to find real world health signals in real world data following time policy changes elsewhere in the world.

A previous well advertised study focusing on transition itself found statistically significant signals for increased heart attacks that magically vanished when averaged over the week. Basically there was a fleeting temporal shift of a few days over when events would occur.

In the absence of compelling evidence and the fact this was already tried in the 70s where this experiment ended in popular rejection I prefer to keep the current policy the way it is.

There are three time zones across the 48 contiguous states.

How does one "opt out"? What does this even mean? Does everyone else now organize their time around your schedule?

There are no "core features" that depend on TPM. Not even bitlocker requires TPM. Personally I do not want and have no use for Microsoft FDE. Ditto for Hello and similar BS.

An encrypted disk provides no useful security benefit and represents an unnecessary additional vector for data loss. The tying of keys to a physical computer .. keys wiped whenever you so much as update bios is a bad idea and a waste of my time.

I use class 0 to protect storage on my laptop against physical theft which has no overhead or OS dependencies whatsoever. I trust this solution far more than bitlocker /w Microsoft's shady auto upload of encryption keys to Microsoft.

Car locks can be easily bypassed. Designing car locks in such a way it becomes meaningfully difficult to bypass serves no useful purpose. People who care will just break the glass to gain entry.

Slashdot is not a monolith neither are individual needs and value judgements. Personally I want security features that protect me from attackers rather than conspiring against me or seeking to protect systems from users. I have a laundry list of useful basic security features Microsoft has spectacularly failed to deliver on.

What sets AI apart from historical precedence is the immortalization of "dead labor". When it can do everything living labor can do and for free the whole technology creating opportunity thing goes out the window. While this is far from what AI currently is a system that lives up to the hype sure as hell would be.

TSA started using naked body scanners / rape scan (e.g. Rapiscan) machines under the Obama administration. I stopped flying after that so for me the data sale thing is moot.

At this point anything is better than the MAGA cult lead by an incompetent treasonous insurrectionist rapist murderer but this isn't saying much. Democrats have a track record of willingly selling out on privacy and liberty.

Password cracking is all about results. 1000x is thermal noise in regimes where exponents protect systems. So a password takes hours instead of minutes or days instead of hours and you have to throw more cores at the problem than you would otherwise. At the end of the day outcomes and expenditures do not meaningfully change.

Your reference is irrelevant, you still don't understand the problem. Hashcat isn't breaking RC4 what it is actually doing is running a dictionary attack. It doesn't matter how shitty RC4 is when the weakest link is entropy not crypto.

See RFC4757 section 2.

"The RC4-HMAC string to key function is defined as follows:
String2Key(password)
K = MD4(UNICODE(password))"

The key to the HMAC or the stream cipher is the goddamn windows password (NTOWF). This means if you take an HMAC or encrypt a known plaintext with this key an attacker is able to brute force the windows password.

From memory the AES mode is still password based only using an amplification scheme so technically it takes some multiplier of additional resources say 1000x or so to accomplish the same task as the RC4 variant. Operationally the difference between RC4 and AES are irrelevant with regards to kerberoasting.

I think this is insane but your choice.

I fail to see the relevance. If not RC4 it would have been AES. Disabling RC4 does NOT solve the problem.

Kerberoasting is independent of RC4. For example hashcat supports brute forcing AES as well. The problem is entropy not algorithm selection.