bradknowles

The cooling system is a major culprit. Especially if they use a zero-waste recycling cooling system. There are all sorts of toxic materials inside the computers being cooled, and those tend to leach out into the cooling fluid in the case of Vera Rubin class processors that must be liquid cooled. And those cooling systems themselves have a lot of toxic materials inside of them, to begin with. So, you have to filter out some of those toxic materials and remove them from the cooling fluid. That’s called “blowdown”.

There are some blowdown processes that leave you with a thick nasty brine that has to be trucked off site. Some blowdown processes take it all the way to creating little dried pucks of super toxic materials, and likewise have to be trucked off site.

If there is a leak of any of these toxic materials, then that can affect nearby creeks, and potentially run off into rivers and lakes that are used for intake of drinking water for cities, in addition to killing or otherwise harming all sorts of local creatures that might be endangered.

Leaks can also potentially leak into the karst geography below the facility, and into the aquifers. Which are also used as inputs to city drinking water sources.

The problem there is that a surface leak might be able to be remediated by scooping up a bunch of poisoned dirt or other materials, and trucking it off site. But you can’t do that with toxic materials that leak into the karst geography and aquifers underground. You can’t just dig up billions of tons of soil and rock to get down to the stuff that’s been poisoned and then remove that.

Likewise, many datacenters have diesel generators on site, and that means they also need diesel fuel storage. If that leaks, it’s also toxic.

Sadly, many datacenters will supposedly “treat” their blowdown and just dump it into the city sewers. Which has about the same effect as leaks.

I’m sure there are other sources to be concerned about, these are just the two main ones I’ve come across so far.

Not yet. I want to publish my skill on GitHub and then let other people contribute ideas to it.

But as it stands today, there’s still too much personalization that is currently built into that skill.

I need to remove that personalization stuff and make it more general. Once I’ve done that, I am planning on publishing it.

For Wall Street, I think they feel like Google/Alphabet is just an advertising company.

Everything else is an also-ran, and doesn’t materially affect anything, unless it costs a huge amount and doesn’t provide a suitable ROI. And here, AI and mega data centers would definitely be a problem — way too much capex, not nearly enough ROI.

Get your own battery systems, that you pay for yourself.

Once they are installed, you can then choose to join a VPP program, where you control how much energy will be sold back, and when.

Of course, you won’t make as much money, but you will have more control over how much power is left in your batteries for when you might really need it.

Reddit will lose any revenue they could have booked from people who were directed to the article but don’t have an account.

Reddit will no longer be searchable by search engines, and that will dry up any source of revenue that they could have gotten from those people who do not have an account.

People aren’t going to bother to sign up for an account just to read a Reddit article.

When you make yourself a completely walled-off society, you will slowly leak people who can’t be bothered to sign in every time. And you can no longer draw in people who do not currently have an account.

That’s a sure-fire recipe for ending yourself. Sooner or later.

I’m sure that Skynet will make good use of this technology.

That’s what he wanted.

But that’s not the only reason he started that war. He also wanted to force the US military to expend all of their munitions, so that they would be forced to buy new ones from his other bestest buddies.

I wonder how much they are paying him for all these free profits?

No, we are not there yet. Right now, AI is de-skilling the engineers who are using it.

Sure, they may churn out a lot more code, but what is the quality of that code? What is the security of that code?

Can you prove that the code they are generating is actually having a material impact on the bottom line?

If the AI tools suddenly became too expensive to use, could those engineers go back to doing things the way they once did?

If you want to vibe code, feel free to do that on your own personal projects. For everything else, you need to be able to demonstrate that what you’re doing is not just fluffing your lines-of-code metrics, or your tokenmaxxing metrics, and actually demonstrate real-world positive impact on your bottom line.

Maybe in a few years, we will get to a point where no one will need those skills anymore, and all that stuff will be done locally, so we don’t need any more massive datacenter projects that suck down all the power and all the cooling and create more problems than they solve.

But we are not there today. And the way that the AI bubble is already crumbling, I’m not convinced that we will get there any time soon.

IMO, a changelog should only be updated once the PR has been checked and validated and gone through all the appropriate CI/CD processes, and finally merged to the main branch. Then you can point at the PR for all their work that was done, with a brief summary in the changelog.

In contrast, if you have ever used SIPRnet, and you have used IP addresses or DNS domains, you’re welcome.

Back in 1995, I was the Domain Technical POC for DISA.MIL, and working in the basement of the Pentagon as a government civilian GS-11. I got wind that the SIPRnet Program Office was planning on using random IPv4 network numbers that they were going to pull out of their @$$. And the same with DNS domains. In fact, they were going to use HOSTS.TXT tables instead. So, I had to track down the person who was the lead program manager and set them straight.

In the end, after an hour-long meeting, the most convincing argument I had was that the NSA was known to be working on a “multi-level secure gateway” that would allow SIPRnet to be connected to the “real live Internet”. It would decide, in real time, what packets were safe to be transmitted in which direction, and which ones were not — with a guarantee of preventing any classified information from ever escaping SIPRnet.

The problem would be that if they randomly chose to use an IPv4 block or a DNS name that already existed in the “real Internet”, then they would be unable to communicate with that endpoint because the packet routing algorithms wouldn’t be able to tell where to route the data.

So, they agreed to go to NIC.DDN.MIL to get all their network numbers assigned to them. And they also agreed to go to NIC.DDN.MIL for all their domains that they wanted to operate. Even though they would be disconnected from the “real Internet” at the beginning, they wanted to be ready to do so once the MLS Gateway came online. Which, of course, never happened. And later, we got other sources for things like network numbers and DNS domains, instead of the one and only NIC.DDN.MIL.

Whew! Barely dodged that bullet. You’re welcome.

Any generalizations that you apply to an entire gender will necessarily be wrong. At least for a large percentage of that gender. Especially if you are not yourself a member of that gender. Same with any other social subgroup.

If you want to talk about all the members of that subgroup that you have personally encountered, that’s a different matter.

It is entirely possible that the handful of women you have met would fall under the descriptions you gave. It is equally likely that there is something about your own personality that tends to cause you to meet women who may tend to fit those generalizations.

Lesson: Don’t ever lump an entire subgroup into a single bucket like that. I guarantee you that there is far more variety out there than you can possibly conceive of, if only you will allow yourself to see them.

I wonder if anyone has done a deep comparison of this package versus the others in this space?

What about benchmarks?

The actual key cannot be exfiltrated from the Secure Enclave. All the encryption/decryption is done inside the Secure Enclave. Knowing the owner’s password is just enough to let you pass data through the Secure Enclave to do the appropriate encryption/decryption operations.

So, you can’t copy that key over to a different Secure Enclave. If you could, then it would no longer be a Secure Enclave.

That’s a wipe and restore type option. Because the RAM and SSD are part of the SOC chip that also includes the CPU cores, the GPU cores, etc….

So, afterwards you effectively have a new iPhone with upgraded storage. Which you then have to restore somehow.

That is, unless you can find someone who can de-lid the SOC, remove the tiny RAM and/or SDD storage chips, and replace them with new ones. But then you’d have the same problem that the data was on the old chips, and encrypted. And so you can’t just copy that data over to the new chips, you have to decrypt all the data using the key stored on the Secure Enclave and copy it out in unencrypted format, then once the old chips are removed from the SOC and replaced, you then have to copy back all that data onto those new chips and have it re-encrypted from the Secure Enclave.

Anyone who can do that could just give you a new iPhone with more storage.

For Apple devices, the storage is usually encrypted. And the encryption key is on the Secure Enclave, and cannot be exfiltrated.

So, if the machine dies, you can’t just take the SSD and solder it onto another board somewhere. Even if you could, the SSD is part of the SOC that includes the CPU, and the Secure Enclave is part of the logic board that it was soldered onto.

Recovering data from an Apple device is intentionally hard to do. Apple wants to protect that device as much as possible from anyone who might try to steal your data by simply unsoldering chips and soldering them back on somewhere else.

Yup. That’s the real lesson that Curtis teaches in his books on the subject.

Your backup is worthless if you haven’t actually tested to confirm that it actually works. And you have to periodically, re-test, as well.

The book “Backup and Recovery” was published on February 6, 2007, almost twenty years ago. I was a technical reviewer for an earlier version of this book, “Unix Backup and Recovery”, published on January 1, 1999. And these books were not the first to advocate the standard 3-2-1 model of backups. Curtis also has more modern books, if you like — such as “Modern Data Protection: Ensuring Recoverability of All Modern Workloads”, published in 2021.

If you’re not adequately backing up your data, then that is entirely on you. After thirty-plus years of this best practice guidance, there is no excuse for that kind of incompetence.

Apple makes this pretty easy with iCloud and Time Machine, but they are far from the only company trying to make this as easy as possible for their customers.

IMO, there is so much more to AGI than what we get today with LLMs. I think we are still at least decades away.

So, there are solutions to pretty much all of these problems. The only issue is that they are expensive.

You can get zero-waste recycling cooling systems, which take in almost no water after the initial fill. The main problem that you then have to deal with is the contaminants that get into the cooling loop and which have to be filtered out — that’s called “blowdown”. There are both liquid and solid blowdown solutions, but in the end that has to be trucked offsite to be dealt with separately.

And in terms of water use, if you’re getting your power off-site, then whoever is generating that power is almost certainly using 5-10x as much water as your datacenter is, and that’s assuming you’re using the most water efficient method of traditional power generation — combined cycle natural gas turbines. Coal and nuclear are even worse. Check the HARC report for more on this.

For power, sure it’s easiest to just hook up to the grid. But you can also do behind-the-meter solutions based on natural gas turbines, renewables like wind or solar, etc…. If you do renewables, then you also need to connect up grid-scale batteries to handle the power draw when the renewables aren’t generating.

Then you have to deal with the fire risk. If you’ve got tons of lithium-ion batteries onsite, or tons of diesel fuel onsite for your diesel generators, then those could easily start a fire. And for lithium-ion batteries, about all the fire companies can do is sit there and wait for them to burn out. That’s when you wish you had done LiFePO4 instead. Once an equipment fire gets going, that can ignite nearby fuel sources, and if you’re close to the WUI at your site, that can get a wildfire started. That could potentially take out hundreds or thousands of nearby houses, schools, other businesses, etc….

Natural gas turbines rarely catch fire, so they’re not the same level of risk as diesel. And natural gas turbines don’t have large fuel tanks onsite, unlike diesel. But you have to be careful with natural gas and ensure that you properly winterize all your pipelines, otherwise you run into the risk of a complete power outage, and the kinds of problems we saw here in Texas during the “Great IcePocalypse”, a.k.a., “Winter Storm Uri”.

For the exhaust of greenhouse gasses, those can be treated before they are released into the atmosphere, but treatment is expensive.

If you make data centers cover all of their costs and not externalize any of those costs onto the people, then they quickly become unprofitable in many places. Which is why all the hyper-scalars building data centers do everything they can to hide everything they’re doing and to buy off all the politicians who could cause any problems for them. Case in point — Texas.

You don’t comprehend what actually happens inside of AWS. 99.999% of the job is dealing with ancient proprietary Amazon systems that were designed a decade or two ago, and have an unbelievable amount of warts and sharp edges. And you’re not allowed to use any newer tools, because Amazon invented “Not Invented Here” a couple of decades before anyone else in the world ever heard of that term.

Yes, they make use of some open-source products as the core of some of the software and services they provide. But as an internal AWS employee, you’re not allowed to use anything but Tier-1 products as components to your software or service, if you use any AWS products or services at all.

This way you don’t have Tier-1 services dependent on Tier-2 or higher services. Priority inversions are avoided at all costs.

But then you still have all the control planes for all of these services which are so heavily dependent on US-East-1. And if US-East-1 goes down, then the data plane has to continue trying to function blindly, because the control plane doesn’t exist anymore. They’re still working on trying to fix that problem.

Open source software, or any software not designed and developed by Amazon/AWS, is no more than 0.001% of the job — if that much.

With respect, when I played pool in a league, that was NAPA. They have an Elo-based system of ranking players, on a nominal 1-100 basis. You can’t go below 1, but you can go above 100 if you’re really good. I’ve seen some 100+ players. They were truly amazing.

However, my best rank was 12. That’s twelve, in case you misunderstand the numbers. I was literally the worst player in the entire league.

I don’t think that there’s anyone out there who is much worse than I am.

So, one of the rules I came up with and put into all of my Claude.md files is that after discovering any piece of information, and before doing anything else, that information has to be written to local file storage. That local file storage is then re-read any time the session is compacted or the main process starts up again. Agents may not need this information, but the top-level process does.

And this is on a per-task basis, so we are not polluting the memory of one top-level Claude process with the data from another.

This method should survive all compaction, and allow me to go back and revisit any previous task that I’ve worked on with Claude.

I also have a rule that any important lessons learned also get recorded into the Claude.md, for the benefit of future tasks.

I choose democracy over billionaires, every single time.

Every single one of them can move to a different country and never, ever again come back. And I would be perfectly happy.

The fine article is paywalled. For alternative versions, see one of the following:

https://newrepublic.com/post/217350/trump-team-arranged-priv... https://www.yahoo.com/news/us/articles/trump-team-arranged-p...

Sounds like the beginning of the end, as far as I can tell.

For me, I don’t stop at the LLMs. I use them to do a deep semantic search of the internet. If all I want is a summary, then that may be as far as I go.

But frequently, I want more than just a summary, so the LLM is just a launching point for me to go do a deep dive on the sites it turns up.

Whereas I was thinking that I should start making the migration to Proton.

So, thanks to everyone for the advance warning!

Their problem is that they don’t really leave you with enough power for real emergencies, while they are hyper focused on doing their energy arbitrage.

See https://epicelectrical.com/what-is-base-power/ for one take on this problem.

For me, I’m happy to have batteries at my home and to voluntarily participate in a VPP, up to certain limits. If there is a predictable bad weather situation coming up, then I want to be able to turn off the VPP and store as much energy locally as I can. And I can only do that if I fully own the battery system that is in place. Tesla doesn’t help me there, neither does Base Power.

That’s not the way Wallstreet thinks. They insist on continuing growth, year over year. Accelerating growth, even.

If you don’t show growth every year, then your stock tanks because you can’t show them the profit that they would make through buying your stock today and then selling it next year.

At least, that’s the way they normally think.

The story goes that Black Friday was named that because it was the break-over point between where most companies were in the red versus when they turned the corner and were now in the black.

At Whole Foods eStore, during the time I worked there, virtually all of their profit came from Thanksgiving. Not before, not after. Just that one week. Christmas was also something of a bump, but the rest of the year operated at a loss.